3.3 Information Asymmetries on Secondary Credit Markets

Transcription

3.3 Information Asymmetries on Secondary Credit Markets
Information Asymmetries on Secondary Credit Markets
DISSERTATION
of the University of St. Gallen,
Graduate School of Business Administration,
Economics, Law and Social Sciences (HSG)
to obtain the title of
Doctor of Philosophy in Management
submitted by
Stefan Morkötter
from
Germany
Approved on the application of
Prof. Dr. Beat Bernet
and
Dr. Simone Westerfeld
Dissertation no. 3676
Difo Druck GmbH
The University of St. Gallen, Graduate School of Business Administration, Economics, Law and Social Sciences (HSG) hereby consents to the printing of the present dissertation, without hereby expressing any opinion on the views herein expressed.
St. Gallen, June 11, 2009
The President:
Prof. Ernst Mohr, PhD
Acknowledgements
I
Acknowledgements
Writing this dissertation was a great experience and a fascinating process to which a lot of
people have contributed in many ways and whom I hereby would like to thank for their diverse support.
First of all I would like to express my deepest gratitude to my thesis supervisor and academic
mentor Prof. Dr. Beat Bernet. Under his guidance I not only accomplished this dissertation
but also started a voyage into the academic world, which I hope will be the beginning of a
long journey. Working at his chair at the Swiss Institute of Banking and Finance at the University of St. Gallen broadened both my academic as well as personal mind-set and was great
experience in several ways.
Furthermore I would like to thank Dr. Simone Westerfeld for having been an inspiring cosupervisor as well as a challenging discussion partner throughout my time at the Swiss Institute of Banking and Finance. Her input and help proved to be invaluable for the development
of this dissertation.
I also own very much gratitude to my friends and colleagues in St. Gallen or back home in
Germany. I cannot name all of them, but I would like to mention Matthias Hoffmann and Andreas-Walter Mattig, with whom I worked together at the Swiss Institute of Banking and
Finance. We all enjoyed a great time and remarkable team work. In addition, I would like to
thank Dr. Alexander Bönner and Roman Frick for spending so much time on fruitful discussions and sharing the experience of being a doctoral candidate.
Part of my thesis was written during my stay at the University of Oxford in England, where I
was cordially welcomed by a great academic community. I specifically want to thank Prof.
Tim Jenkinson, who always had an open door for me and offered help and advice throughout
various valuable discussions. Furthermore, I thank the Swiss National Science Foundation
(SNF) for financial help.
Last but not least, I want to thank my parents, since I could rely on them throughout my
whole life. Without their unconditional support and care this dissertation would never have
been possible. I dedicate this dissertation to them.
St. Gallen, April 2009
Stefan Morkötter
Contents
III
Contents
Acknowledgements ........................................................................................................ I
Contents .......................................................................................................................III
Abstract ...................................................................................................................... VII
Abstract in German .................................................................................................... IX
Abbreviations ........................................................................................................... XIII
List of Figures ........................................................................................................... XVI
List of Tables .......................................................................................................... XVII
1 Introduction ............................................................................................................. 1
2 Research Set Up ....................................................................................................... 5
2.1 Guiding Research Questions ............................................................................ 5
2.2 Research Topic ................................................................................................. 7
2.3 Research Object ............................................................................................. 14
2.4 Scientific Objective ........................................................................................ 18
3 Information Asymmetries on Secondary Credit Markets ................................. 20
3.1 Information Asymmetries .............................................................................. 20
3.1.1 Basics ...................................................................................................... 20
3.1.2 Quality Uncertainty ................................................................................ 22
3.1.3 Moral Hazard .......................................................................................... 22
3.1.4 Hold-Up .................................................................................................. 23
3.2 Information Agents on Secondary Credit Markets ........................................ 23
3.2.1 Credit Rating Agencies ........................................................................... 23
3.2.2 Stock Analysts ........................................................................................ 25
3.3 Information Asymmetries on Secondary Credit Markets .............................. 27
3.3.1 Issuer and Investor .................................................................................. 27
3.3.1.1 Research Subset I ................................................................................ 27
3.3.1.2 Research Subset II............................................................................... 28
3.3.2 Investor and Information Agent ............................................................. 29
3.3.2.1 Research Subset I ................................................................................ 29
3.3.2.2 Research Subset II............................................................................... 29
3.3.3 Issuer and Information Agent ................................................................. 30
3.3.3.1 Research Subset I ................................................................................ 30
3.3.3.2 Research Subset II............................................................................... 31
4 Rating Model Arbitrage in CDO Markets: An Empirical Analysis ................. 32
4.1 Introduction .................................................................................................... 32
4.2 Literature Review ........................................................................................... 34
IV
Contents
4.3 Information Asymmetries within CDO Markets and the Role of CRAs ....... 37
4.4 Data Sample ................................................................................................... 42
4.5 Empirical Results ........................................................................................... 44
4.5.1 Univariate Tests ...................................................................................... 44
4.5.1.1 Set I (sorting by Rating Agencies) ...................................................... 45
4.5.1.2 Set II (sorting by Rating Methodologies) ........................................... 49
4.5.2 Multivariate Tests ................................................................................... 50
4.5.2.1 Set I (sorting by Rating Agencies) ...................................................... 51
4.5.2.2 Set II (sorting by Rating Methodologies) ........................................... 54
4.5.3 Interpretation........................................................................................... 58
4.6 Conclusion...................................................................................................... 60
5 Impact of Multiple CDO Ratings on Credit Spreads ........................................ 62
5.1 Introduction .................................................................................................... 62
5.2 Literature Review ........................................................................................... 63
5.3 Multiple Ratings and Credit Spreads within CDO Markets .......................... 66
5.4 Data Sample ................................................................................................... 69
5.5 Empirical Results ........................................................................................... 72
5.5.1 Analysis of Credit Spreads ..................................................................... 72
5.5.2 Impact of Multiple Ratings ..................................................................... 78
5.5.3 Decreasing Reduction of Underlying Tranche Spreads ......................... 79
5.5.4 CDO Tranches rated by Fitch ................................................................. 81
5.5.5 Regression Analysis................................................................................ 83
5.6 Conclusion...................................................................................................... 90
6 Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS
Spreads ................................................................................................................... 93
6.1 Introduction .................................................................................................... 93
6.2 Literature Review ........................................................................................... 95
6.3 Spill-over Effects between Stock Analysts’ Forecasts and CDS Spreads ..... 99
6.3.1 Mean Stock Analysts’ Forecasts and CDS Spreads ............................... 99
6.3.2 Dispersion of Mean Stock Analysts’ Forecasts and CDS Spreads....... 102
6.4 Data Sample ................................................................................................. 104
6.5 Empirical Results ......................................................................................... 109
6.5.1 Co-movement ....................................................................................... 109
6.5.2 Lead-lag Structures ............................................................................... 113
6.5.3 Long-run Equilibrium Relationship...................................................... 119
6.6 Conclusion.................................................................................................... 122
7 Conclusion and Outlook ..................................................................................... 125
7.1 Summary of the Results ............................................................................... 125
7.2 Relevance for Market Participants and Regulatory Authorities .................. 132
Contents
V
References ................................................................................................................. XIX
Appendix ...............................................................................................................XXXV
Curriculum Vitae ............................................................................................ XXXVIII
Abstract
VII
Abstract
In the course of the current financial crisis specifically the valuation of assets with an
exposure to secondary credit markets has not only become problematic due to the
eroding value of the underlying collateral pools but also due to a lack of trust between
market participants. Lack of trust is associated with low levels of transparency that is
also impacting information asymmetries between market participants. The dissertation’s aim is to provide empirical evidence for the existence of information asymmetries on secondary credit markets along three different empirical settings. Secondary
credit markets are defined as a market place for resale activities of credit-linked assets.
First, the dissertation investigates the CDO rating process and shows that information
asymmetries between the issuer and the investor (may) lead to rating model arbitrage.
In this context rating model arbitrage is defined as the issuer’s deliberate capitalization
on information asymmetry at the investor’s cost on the basis of different rating
processes as applied by the rating agencies.
Second, it is analyzed whether multiple ratings for CDO tranches have an impact on
credit spreads. Against the background of information asymmetries the dissertation
therefore examines various effects with regard to the number of rating agencies involved. The results indicate foremost a negative correlation structure between the
number of outstanding ratings and the corresponding credit spread levels.
Third, the dissertation discusses the dynamic relationships between stock analysts’
earnings forecasts and CDS spreads. It is shown that higher forecasts are associated
with lower CDS spreads, whereas the dispersion of analysts’ forecasts is positively
correlated with CDS spread levels. Significant lead-lag structures are detected between
the dispersion of analysts’ forecasts and CDS spreads with the latter leading the first.
Empirical evidence of information asymmetries on secondary credit markets in turn
helps regulatory authorities to track down problem areas. Once the critical issues are
identified, future regulatory standards can be defined in order to increase financial stability and revitalize activities on secondary credit markets. Thus, the dissertation can
also be seen as a practical contribution, as it adds to the understanding of the current
financial crisis from an empirical angle and draws respective conclusions.
Abstract in German
IX
Abstract in German
Im Zuge der aktuellen Finanzmarktkrise wurde die Bewertung von Kapitalanlagen auf
Sekundärmärkten für Kreditrisiken vor dem Hintergrund mangelnden Vertrauens zwischen den Marktteilnehmern zunehmend schwierig. Der eingetretene Vertrauensverlust ging einher mit niedriger Markttransparenz und dadurch bedingt mit Informationsasymmetrien zwischen den jeweiligen Marktteilnehmern. Vor diesem Hintergrund
hat die vorliegende Dissertation das Ziel, anhand von drei unterschiedlichen Untersuchungen den empirischen Beweis für Informationssymmetrien auf Sekundärmärkten
für Kreditrisiken zu erbringen. Sekundärmärkte für Kreditrisiken werden im Folgenden definiert als Marktsegment für den Handel von kreditgebundenen Kapitalanlagen.
In einer ersten empirischen Untersuchung werden die Ratingprozesse im Rahmen von
CDO Transaktionen näher betrachtet. Es wird aufgezeigt, inwieweit Informationsasymmetrien, die aus dem Dialog des Emittenten mit der Ratingagentur erwachsen, zu
Rating Model Arbitrage führen können. Rating Model Arbitrage bedeutet, dass der
Emittent auf Kosten der Investoren die bestehenden Informationsasymmetrien zum
eigenen Vorteil nutzt und die Transaktionsstruktur dahingehend optimiert.
Vor dem Hintergrund einer zweiten empirischen Analyse wird der Einfluss untersucht,
den die Anzahl ausstehender Ratings einer CDO Tranche auf den zugrunde liegenden
Zinssatz hat. Eine wesentliche Erkenntnis in diesem Zusammenhang ist, dass die Anzahl der ausstehenden Ratings negativ mit dem Zinssatz der CDO Tranche korreliert
ist: Je mehr Ratingagenturen eine Tranche bewerten, umso geringer ist der Zinssatz,
den ein Investor für seinen Kapitaleinsatz erhält.
Abschliessend untersucht diese Dissertation übergreifende Effekte zwischen Konsensschätzungen von Aktienanalysten und CDS Märkten. Es zeigt sich, dass höhere Konsensschätzungen von Aktienanalysten mit niedrigen CDS Spreads verbunden sind.
Hinsichtlich der Interaktion von CDS Spreads mit der Streuung von Analystenschätzungen wird darüberhinaus deutlich, dass die Streuung von Analystenschätzungen den
CDS Spreads nachfolgt.
Der empirische Nachweis von Informationsasymmetrien ermöglicht Aufsichtsbehörden im Zuge der aktuellen Finanzmarktkrise, die Marktstandards auf Sekundärmärkten
Abstract in German
XI
für Kreditrisiken möglichst zielführend zu überarbeiten. Dies wiederum sollte die Finanzmarktstabilität erhöhen und zukünftige Emissionsaktivitäten auf den Sekundärmärkten für Kreditrisiken positiv beeinflussen.
Abbreviations
XIII
Abbreviations
ABS
Asset-backed Securitizations
bn
billion
bps
basis points
CBO
Collateralized Bond Obligation
CDO
Collateralized Debt Obligation
CDS
Credit Default Swap
CEO
Chief Executive Officer
CESR
Committee of European Securities Regulators
CLO
Collateralized Loan Obligation
Coeff.
Coefficient
CRA
Credit Rating Agency
CSO
Collateralized Swap Obligation
Diss.
Dissertation
e.g.
for example
Ed.
Editor
EL
Expected Loss
EPS
Earnings-per share
ESME
European Securities Markets Expert Group
et al.
et alii
etc.
et cetera
EU
European Union
EUR
Euro
FEPS
Mean Stock Analysts’ Forecasts of Earnings-per-share
Fitch
Fitch Ratings
Abbreviations
XIV
GDP
Gross Domestic Product
I/B/E/S
Institutional Brokers Estimate System
IOSCO
International Organization of Securities Commissions
ISDA
International Swaps and Derivative Association
JCF Group
Jacques Chahine Finance Group
KS Test
Kolmogorov-Smirnov-Test
LGD
Loss given Default
LIBOR
London Interbank Offered Rate
LSTA
Loan Syndications and Trading Association
m
million
M&A
Mergers and Acquisitions
Moody’s
Moody’s Investors Service
o.P.
other Places
p.
page
PD
Probability of Default
pp.
following pages
S CDO
Synthetic Collateralized Debt Obligations
S&P
Standard & Poor’s
SD
Standard Deviation
SEC
Securities and Exchange Commission
SME
Small-and-medium sized Companies
SPV
Special Purpose Vehicle
U.S.
United States
USD
US-Dollar
VAR
Vector Auto Regression
Abbreviations
XV
vs.
versus
www
world wide web
List of Figures
XVI
List of Figures
Figure 2-1: Structure of a Credit Default Swap .............................................................. 9
Figure 2-2: Structure of a Collateralized Debt Obligation ........................................... 11
Figure 2-3: General Research Object and the Corresponding Subsets ......................... 16
List of Tables
XVII
List of Tables
Table 4-1: Comparison of Subgroups (Set I & II) ........................................................ 45
Table 4-2: Test of Equality of Group Means - Set I (ANOVA) ................................... 46
Table 4-3: Kolmogorov-Smirnov- Test (Set I & II) ..................................................... 48
Table 4-4: Test of Equality of Group Means - Set II (ANOVA).................................. 49
Table 4-5: Discriminant Analysis and Classification of Set I ...................................... 51
Table 4-6: Discriminant Analysis of Set I (sorting by Rating Agencies) ..................... 52
Table 4-7: Classification of Results of Discriminant Analysis V................................. 54
Table 4-8: Discriminant Analysis and Classification of Set II ..................................... 55
Table 4-9: Discriminant Analysis of Set II (sorting by Rating Methodologies) .......... 57
Table 5-1: Data Sample “Multiple CDO Tranche Ratings” ......................................... 70
Table 5-2: Mapping Code for the Individual Rating Notches ...................................... 71
Table 5-3: Notch Differences of jointly-rated CDO Tranches ..................................... 72
Table 5-4: Credit Spread of CDO Tranches (Multiple Rating and Rating Code) ........ 76
Table 5-5: Credit Spread of CDO Tranches (Rating Agency and Rating Code) ......... 77
Table 5-6: Robustness Checks for the Grouping Factor Multiple Ratings ................... 78
Table 5-7: Multiple comparisons of underlying tranche spread differences ................ 81
Table 5-8: Comparison of Rating Outcomes (Rating Agencies and Rating Code) ...... 82
Table 5-9: Impact of Multiple Ratings (Multiple Regression Analysis) ...................... 88
Table 5-10: Robustness Checks (Controlling for Size Effect) ..................................... 89
Table 6-1: Overview of Mean Analysts’ Forecasts and CDS Spreads ....................... 107
Table 6-2: Analysis of Correlation ............................................................................. 112
Table 6-3: Panel Data Analysis (Fixed Effect Model) ............................................... 115
Table 6-4: Vector Auto Regression Analysis (Lag Structures) .................................. 118
XVIII
List of Tables
Table 6-5: Long-run Equilibrium Relationship between CDS Spreads and Stock
Analysts’ Forecasts .............................................................................................. 122
Introduction
1
1 Introduction
In the course of the current worldwide financial crisis specifically the valuation of asset securitizations is a key aspect and has led the capital markets to a new stage of escalation only comparable to the Great Depression in the thirties of the last century.
Valuation of asset securitization has not only become problematic due to the eroding
value of the underlying collateral pools but also due to a lack of trust between market
participants (Krinsman, 2007). Lack of trust in turn has led to a drain of liquidity, making it impossible to obtain mark-to-market valuation for a multitude of asset securitizations. Ultimately, the loss of confidence between market participants has not been limited to secondary markets and subprime debt only but in the ensuing months it has
affected large parts of the international financial intermediation system ranging from
interbank markets and even to retail banking activities. Against this background Josef
Ackermann, CEO of Deutsche Bank, justly labeled current market turmoil as a crisis
of confidence (Deutsche Bank, 2008).
In addition, lack of trust within the affected financial markets can also be viewed as a
function of opaqueness. Low levels of transparency lead to unequal information distribution or information asymmetries respectively between market participants harming
the process of optimal asset allocation. Often information asymmetries are not coincidentally taking place, but are a systematic appearance and merely a result of specific
market regimes. One way to overcome the current crisis of confidence is therefore an
approach that reduces existing information asymmetries and increases levels of transparency. However, prior to potential counter measures, relevant information asymmetries need to be detected. With regard to secondary credit markets the issue of information asymmetries has yet not been analyzed in much detail from an empirical angle.
The research objective of this dissertation is therefore to provide empirical evidence
for the existence of information asymmetries on secondary credit markets along three
different empirical settings.
After the financial markets seemed to overcome a first wave of market turmoil in late
2007 or early 2008, respectively, the discussions immediately centered on the issue of
who to blame for what was then primarily still being referred to as a subprime crisis.
2
Introduction
Increasingly, rating agencies found themselves, their business model and their rating
processes for assessing asset securitizations under heavy attack, not only from academics, politics, and market participants but also from the international media (e.g. Ashkraft and Schuermann, 2008, pp. 5; Chirico, 2008; ESME, 2008; Lucas et al., 2008;
Manns, 2008; SEC, 2008; Sinclair et al., 2008; Zuberbühler, 2008). Rating agencies
were blamed for not downgrading the involved subprime securitizations precociously
and thus holding onto exaggerated ratings for too long. The existing practice of delegating monitoring activities in the course of structured finance investments to rating
agencies has not proved to be sustainable enough from the investors' perspective (e.g.
Borio, 2008; Milleker and Sauerschell, 2008). Mandated by the European Commission, the very recent Larosière Report (2009) explicitly listed the failure of credit rating agencies as one of the main causes leading towards the financial crisis. With the
financial crisis intensifying from March/April 2008 the discussion relating to the role
of rating agencies began to focus on regulatory issues that are likely be lead to increased supervision of rating agencies through government authorities in the future. In
particular, the widespread perception that rating agencies not only provide the financial community with ratings but also offer consulting services to the issuer for the very
same transaction was criticized (e.g. Buiter, 2007; CESR, 2008a, 2008b; IOSCO,
2008). This argument correlates with a growing demand for higher levels of transparency both relating to the rating process as well as to secondary credit markets in general, with the latter explicitly including credit default swaps (CDS) (e.g. Cox, 2008).
Most notably, the European Union put forward several proposals that sought to reinforce the supervision of rating agencies and would result in a clear separation between
consulting and rating services (EU Commission, 2008).
Due to extensive on- and off-balance sheet activities worldwide contagion observed
throughout the very recent financial crisis is also a function of closely interlinked financial institutions. Given very recent occurrences it is thus rather unlikely that effects
triggered by a financial crisis are limited to a specific local area (country) or market
niche at present. Regulatory authorities are thus faced with a complete new quality of
financial crisis demanding both quick and coordinated countermeasures. Increasing
worldwide interaction in the financial sector is also closely associated with the rise of
Introduction
3
secondary credit markets and the underlying asset classes and (hybrid) investment
structures. With regard to secondary credit markets, the financial crisis that began in
2007 is the first stress test under real market conditions. Secondary credit markets
mainly consist of structured finance products and credit derivatives and thus focus on
resale activities of credit-linked assets on a broad scale. Two important examples of
secondary credit markets are collateralized debt obligations (CDOs) and CDSs. For
these comparatively young asset classes the years preceding the financial crisis were
characterized by indefinite growth accompanied by rather few drawbacks. Supported
by exuberant liquidity dynamics and low interest rates for plain vanilla bonds, market
participants (e.g. investment banks) primarily focused on the growth of issuance levels
rather than on the development of regulatory initiatives and mature settlement mechanisms. CDO issuance volumes for example increased from USD 59.0bn in 2002 up to a
record level of USD 343.0bn in 2006 followed by USD 302.4bn in 2007 (Hu, 2008;
Rajendra at al., 2008; S&P, 2008). However, in terms of their structural properties
both CDS and CDO markets remained on a level defined in an early stage of market
evolution. The market architecture therefore primarily relies on an over-the-counter
business model, which in turn correlates with high levels of information asymmetries
and low transparency in general.
Up to now, both academics and practitioners have for the most part argued from a rather limited empirical perspective with regard to the existence of information asymmetries on secondary credit markets inducing the above stated crisis of confidence. They
have seldom attempted to actually spot and address information asymmetries on secondary credit markets from an empirical perspective. In the past, financial literature
focused on valuation models of secondary credit markets products and not so much on
information asymmetries (e.g. Hull and White, 2004; Longstaff and Rajan, 2008). This
research focus seems fairly understandable, given the fact that the financial industry
was in need of proper pricing models that would allow the trading of new products
(e.g. CDS or CDOs). However, recent circumstances not only exposed the limitations
of the existing models but have also drawn more attention to the causes of the financial
crisis, including research on information asymmetries (e.g. Duffie et al., 2009).
4
Introduction
Based on these considerations, the dissertation’s research objective is to provide empirical proof of information asymmetries on secondary credit markets with a particularly focus on CDS and CDO markets. Against the background of three different empirical settings the dissertation aims to uncover systematic failures with respect to unequal information distribution between participants on secondary credit markets.
Throughout the different empirical settings the dissertation’s research objective centers
on the relationship and interaction between different market participants specifically
aiming at the role of information agents (e.g. rating agencies). The targeted research
initiatives put a strong emphasis on CDOs and the underlying rating(s) (processes) as
well as on spill-over effects between CDS spreads and stock analysts.
Empirical evidence of information asymmetries on secondary credit markets in turn
helps regulatory authorities to track down problem areas. Once the critical issues are
identified, future regulatory standards can be defined in order increase financial stability and revitalize activities on secondary credit markets. Thus, the dissertation can also
be seen as a practical contribution, as it adds to the understanding of the current financial crisis from an empirical angle and draws respective conclusions.
The dissertation is organized as follows: Chapter 2 comprises a detailed presentation
of the dissertation's research set-up including the guiding research questions, the research topic, the research object as well as the scientific objective. Along the defined
research set up chapter 3 introduces the concept of information asymmetry and applies
it to secondary credit markets. In the chapters 4 to 6 the idea of information asymmetries on secondary credit markets is addressed from an empirical angle. Chapter 4 focuses on information asymmetries leading to rating model arbitrage on CDO markets.
Multiple CDO ratings and its’ impact on credit spreads are assessed throughout chapter 5, followed by an analysis of spill-over effects and information distribution between CDS markets and stock analysts’ forecasts (chapter 6). Chapter 7 concludes the
dissertation.
Research Set Up
5
2 Research Set Up
2.1 Guiding Research Questions
The dissertation’s research objective is to empirically prove the existence of information asymmetries on secondary credit markets. The dissertation focuses on CDO
and CDS markets and makes an analysis along three different empirical settings (i)
if information asymmetries exist, (ii) how information asymmetries impact pricing
structures and (iii) if information asymmetries also exist between secondary credit
markets and other adjacent financial markets.
The dissertation’s research objective is to empirically prove the existence of information asymmetries on secondary credit markets. In the following the dissertation focuses
on CDS and CDO markets and makes an analysis along three different empirical settings (i) if information asymmetries exist, (ii) how information asymmetries impact
pricing structures and (iii) if information asymmetries also exist between secondary
credit markets and other adjacent financial markets. Each of these research perspectives will be addressed in the course of an in depth empirical analysis that follows a
guiding research question:
(I)
Do information asymmetries between issuers and investors lead to rating
model arbitrage in CDO markets? In this context rating model arbitrage is
defined as the issuer’s deliberate capitalization on information asymmetry at
the investor’s cost on the basis of different rating processes.
(II)
Do information asymmetries impact credit spreads of CDO tranches through
multiple CDO ratings?
(III)
Do information asymmetries between stock analysts’ earnings forecasts and
CDS spreads lead to spill-over effects between two adjacent financial markets?
6
Research Set Up
These questions all have in common that they challenge the issue of information distribution on either CDS or CDO markets. Even if each individual research question
focuses on a specific issue under the rather broad general framework of information
distribution on secondary credit markets, the research questions are still interlinked
with each other. By addressing the existence of information asymmetries on CDO
markets and its impact on the behavior of market participants, the first research question lays the theoretical and empirical ground for the following investigations. Thus,
building on the findings of the first research question, the second empirical analysis
goes one step further and analyses the effects of information asymmetries on the pricing structure of CDO credit spreads. The third question finally leaves the intra-market
perspective and challenges the idea of inter-market information asymmetries. By comparing different asset classes and corresponding financial market places (CDS markets
vs. stock analysts) the third question indicates that information asymmetries are not
bound to the level of individual market segments only, but may in addition also exist
between secondary credit markets and other financial market places or their participants respectively. Information asymmetries arising between CDS markets and stock
analysts’ forecasts should eventually lead to spill-over effects between these two adjacent market places. In addition to the exchange of information within one market segment the research focus is thus extended to the functioning of information exchange
modes between two financial market places.
In the course of this dissertation each guiding research questions will be addressed
throughout chapters 4 to 6 through an individual empirical analysis.
Research Set Up
7
2.2 Research Topic
The research topic of this dissertation is secondary credit markets in general and
CDO as well as CDS markets more specifically. Secondary credit markets are defined as a market place for resale activities of credit-linked assets and can be further
differentiated into direct resale activities, derivative resale activities and structured
resale activities. CDOs belong to the segment of structured resale activities whereas
CDS belong to derivative resale activities.
In the following we refer to the research topic as the general outtake of a broader reality around which the research guidelines are centered (Bernet, 2003, p. 3). For this dissertation the research topic is defined as secondary credit markets. Credit markets are
not only by far the biggest asset class in terms of absolute values but can also be subdivided into different segments starting with the origination and issuance of loans and
corporate bonds. This market segment can be defined as the primary credit market. In
addition, particularly over the last years various new credit markets have emerged
which can be aggregated under the segment of secondary credit markets. Since credit
markets are a rather broad asset class with various instruments and different market
segments, the research topic is by intention limited to the fields of secondary credit
markets only. Secondary credit markets are defined as the market place for resale activities of credit-linked assets. Credit-linked assets in turn are financial instruments
whose values are directly or indirectly derived from credit-sensitive assets (Bielecki
and Rutkowski, 2002, p. 169). Besides plain vanilla credit facilities (e.g. loans and
corporate bonds) credit-linked assets include structured, synthetic and derivative instruments. Particularly in recent years secondary credit markets have gained some
prominence (Watson and Carter, 2006, p. xi).
The primary credit market on the other hand centers on the origination of credit facilities. In the case of loans these activities correspond to the lending and origination
process. For (corporate) bonds, primary credit markets are defined as the original issuance through which the borrower collects capital against his balance sheet. This
original bond issuance has to be separated from secondary credit market activities in
which bond offerings might be re-packaged through the application of structured credit
8
Research Set Up
derivatives (e.g. Gorton and Souleles, 2005; Manns, 2008). A bank following a buyand-hold approach would thus limit its activities on managing its credit portfolio to
primary credit markets only. The primary market for bonds mainly consists of bond
offerings undertaken by corporate, governments or government-like institutions.
In addition, secondary credit markets can be grouped into the following sub-segments:
x Direct Resale Activities (e.g. Corporate Bonds)
x Derivative Resale Activities (e.g. Credit Default Swaps)
x Structured Resale Activities (e.g. Collateralized Debt Obligations)
Direct Resale Activities (e.g. Corporate Bonds)
Direct resale activities stand for secondary purchases of loans and bonds after their issuance or origination respectively. Depending on the chosen credit facility, secondary
bond markets are rather liquid. This is particularly true for the secondary market of
large government bonds of high credit quality and involving large issuance volumes.
Secondary loan markets (also labeled as secondary syndication, e.g. Westerfeld, 2008)
are still in a very early evolutionary stage. However, initiatives have recently been undertaken to increase market liquidity for secondary loans (e.g. LSTA, 2008).
Derivative Resale Activities (Credit Default Swaps)
Derivative resale activities comprise credit derivatives. Credit derivatives are financial
innovations that rely on well-known basic derivative structures like swaps or options
(Hull, 2006, pp. 149). In comparison to stock-based equivalents, valuation of credit
derivatives is derived from the credit risk of underlying entities issuing debt, like companies, without directly owning a stake in the underlying (e.g. Bielecki and Rutkowski,
2002, p. 8; Jorion, 2007, pp.454). The value of a credit derivative is aligned to the credit-sensitive performance and not the ownership of an underlying reference entity. The
reference entity in turn can take varies shapes and is not necessarily restricted to a single name perspective only. In addition, credit derivatives allow credit risk to be detached from market risk and transferred to an investor (Chacko et al., 2006, pp. 147).
Research Set Up
9
As the family of credit derivatives includes a range of different instruments both structure and underlying may vary accordingly. Besides CDSs another example for credit
derivatives are credit spread options (O’Kane, 2001, p. 39). However, in terms of liquidity CDSs are among the most actively traded credit derivatives and represent
around half of the whole market for credit derivatives (e.g. ISDA, 2008; Hager, 2008).
Besides CDOs, CDS are the second credit risk instrument analyzed in detail along the
research set-up of this dissertation in order to assess information asymmetries on secondary credit markets. It is therefore explained in more detail in the following.
According to Taylor (2007, p. 149) a CDS is a contract between two parties which allows the ”transfer [of] the credit risk of a reference entity from one party to another“.
A typical reference entity for a single-name CDS is a specific company (e.g. UBS). As
outlined in Figure 2-1 the two contracting parties are characterized as protection buyer
and protection seller. The protection seller guarantees his counterpart protection in
case of a credit event (Norden, 2004, p. 20).
Figure 2-1: Structure of a Credit Default Swap
Premium
Protection
Seller
No Credit Event
Credit Event
No Payment
Protection
Buyer
Repayment
Interest
Credit
Payment
Reference
Entity
Source: own illustration adapted from Effenberger (2004).
Thus, the CDS is comparable to a (credit) insurance model. In exchange for the guarantee the protection seller periodically receives a premium from the protection buyer
on the basis of a CDS. Typically, this premium is a fixed annual spread quoted in basis
points of the notional amount underlying the swap contract (Martin et. al., 2006, pp.
24). As a basic principle the premium received by the protection seller is negatively
correlated with an increasing credit quality of the underlying reference entity. In order
10
Research Set Up
to avoid misunderstandings both parties have to agree on a loan or bond with particular
maturity functions as the underlying of a CDS (Taylor, 2007, p. 155). The protection
buyer has to pay the premium in any event, whereas the protection seller only has to
pay in the case of a credit event. Both the assessment of the credit event as well as
modes of payment in the case of a credit event typically take place according to specific rules defined by the International Swaps and Derivatives Association (ISDA).
In the CDS contract described above it is assumed that the protection buyer is indeed
invested in debt of the reference entity. Thus, the protection buyer hedges his credit
risk position through the application of a CDS. This constellation is called a covered
CDS. However, if on the other hand the protection buyer does not hold a corresponding debt position but nevertheless is engaged in CDSs, it is referred to it in the following as an uncovered CDS. In this context the buyer’s motivation is speculative since he
bets on a decrease in the credit quality of the reference obligation. Speculative intensions can also be found on the protection seller’s side since he bets on the credit quality remaining constant or increasing (Taylor, 2007, p. 157).
Structured Resale Activities (Collateralized Debt Obligations)
Structured resale activities are defined in the following as the whole range of assetbacked securitizations (ABSs) including mortgage-backed securities, CDOs as well as
other transaction structures (e.g. Gregory, 2004, pp. 151; Morr et al., 2005). Since the
dissertation’s research topic exclusively focuses on CDO markets in the area of structured resale activities, the following illustration is by intention limited to CDOs only.
However, the very basic transaction structure as defined in the case of CDOs is also
applicable to other types of ABS transactions.
The securitization process of a CDO centers on the structuring and resale of credit facilities – like loans to small-and-medium sized companies, emerging market bonds or
even tranches of other ABS transactions (Lucas et al., 2007, p. 3). Figure 2-2 outlines
the structure of a plain vanilla true sale CDO transaction. In the course of a true sale
CDO, legal ownership of a pool of credit assets is transferred through the set up of a
special purpose vehicle (SPV) to the investor (e.g. Goodman and Fabozzi, 2002, pp.
15; Duhon, 2006, pp. 126). Throughout the transaction the SPV issues tranches (or
Research Set Up
11
Figure 2-2: Structure of a Collateralized Debt Obligation
Administration
Trustee/Servicer/Calculation Agent
Capital
Capital
Senior Tranche
(Class A)
AAA
SPV
Loan/ Bond
Portfolio
(Collateral)
Transfer of
Assets and Interest
Interest/
Capital
Swap Counterparty
(Hedging of Interest Rate and Currency Risk)
Junior Tranche
(Class B)
AA
Junior Tranche
(Class C)
BBB
Equity Share
Source: Pawley (2004).
synonymously, notes), which in turn are bought by a range of different investors (e.g.
pension funds, hedge funds or other banks). The SPV uses the capital proceeds to invest into a collateral pool. If the collateral pool consists of the very same credit assets
throughout the whole maturity of the transaction and is only reduced by repayments
and amortization, the transaction is described as static. In contrast, if proceeds of repayments and amortization are reinvested or assets of the collateral pool are sold and
new bonds or loans are bought in exchange, the CDO follows a managed approach
(Carter et al., 2006).
Typically, an (investment) bank or an external asset manager acts as an issuer, setting
up the whole transaction structure including the SPV, hiring third parties (e.g. trust
companies or other investment banks) to perform additional administrative services
(e.g. acting as a calculation agent) and eventually orchestrating the tranche issuance
(e.g. Lindtner, 2006, p. 19; Clancy, 2006, p. 40). In the financial literature the issuer is
also denoted as an originator if he is the original creditor of the transferred assets. In
the case of a balance sheet CDO the transaction’s purpose is to transfer CDO-able assets from the issuer’s and the originator’s balance sheet to the SPV, thus freeing capital in order to reduce, for example, required regulatory or economic capital. Arbitrage
deals on the other hand are merely constructed by asset managers in order to benefit
from arbitrage opportunities by arranging a collateral pool and reselling it to other investors (Lucas et al., 2006, p. 9).
12
Research Set Up
As outlined before, the SPV finances its asset purchases through the issuance of
tranches (Ford, 2006, pp. 99). Each tranche pays interest to the investor as well as capital repayment at maturity. Interest payments are either defined as a fixed rate or as a
spread premium over a certain reference benchmark (typically some sort of LIBOR).
Capital and interests of the notes are guaranteed through the asset pool held by the
SPV. Prior to the issuance each tranche is typically rated by Fitch, Moody's and S&P.
The number of rating agencies assigned by the issuer to a transaction varies between
one and three. A specific characteristic of a CDO transaction is that interest payments
and capital repayment of the notes take place in the form of a so-called waterfall, starting with the most senior notes. In very simple terms this waterfall structure can be described as follows: the more subordinated notes only receive interest payments or capital repayments, respectively, if the claims of the senior tranches are satisfied (Duhon,
2006, p. 127). Typically, more senior tranches not only receive higher credit ratings
but are also linked with greater volumes. In financial literature SPV refinancing is perceived to follow a specific pattern described as a seniority structure (e.g. Chan-Lau
and Lu, 2006). In this context, subordinated tranches act as credit enhancement on behalf of the more senior notes. In addition, it is possible to grant liquidity enhancements
through a capital cushion (Bär, 1997, pp. 30). Finally, the structure is completed by a
swap counterparty, allowing the SPV to hedge against interest rate and currency risks.
Typically, other banks or even the originator acts as a swap counterparty (Bluhm and
Overbeck, 2005, p. 121).
CDOs also exist as synthetic structures consisting of a reference pool whose legal
ownership is not transferred to the SPV and remains with the originator. Through the
application of credit derivatives (e.g. CDS) only the credit risk is transferred from the
originator to the SPV and from there to the investor (e.g. Longstaff and Rajan, 2008;
Tavakoli, 2003, pp. 17). Besides synthetic or true-sale-only transactions hybrid structures also exist which consist of both the application of credit derivatives as well as
investments into a collateral pool (e.g. Jobst, 2006). Depending on the underlying asset
(collateral) pool, the individual CDO belongs to one of the following transaction types:
Collateralized Bond Obligations (CBO), Collateralized Loan Obligations (CLO), Collateralized Swap Obligations (CSO) or other (exotic) transactions (Westerfeld, 2008).
Research Set Up
13
For a detailed illustration of the different transaction types, it is at this point referred to
Jortzik (2005, pp. 53), Lucas et al. (2007, pp. 71) and Schiefer (2008, pp. 187).
14
Research Set Up
2.3 Research Object
CDO and CDS markets are analyzed from the perspective of information asymmetries. The research object is based on a set of different principal-agent relationships
which are derived from a triangle of market participants including the issuer, the information agent as well as the investor. Through the introduction of two different
subsets this general research object is aligned to the specific features of CDO as
well as CDS markets respectively. In the case of CDOs, rating agencies perform the
role of information agents, whereas this function is undertaken by stock analysts for
the subset focusing on CDS markets. Thus, in the second subset the relationship triangle opens up for participants of an adjacent financial market and allows empirical
analyses of spill-over effects between CDS markets and stock analysts’ forecasts.
Following Bernet (2003, p. 3) the research object determines the perspective from
which the research topic is analyzed and to which the central arguments of the thesis
are aligned. Accordingly, the dissertation’s research object corresponds to information
asymmetries on secondary credit markets and can be viewed as the central argument
underlying the overall research perspective. Against the background of two subgroups
referring to CDO as well as CDS markets, respectively, various settings of information
asymmetry are analyzed. Chapter 3 provides us with a more theoretical view of the existence of information asymmetries on secondary credit markets, whereas chapters 4 to
6 present the issue of information asymmetries from an empirical perspective. Relying
on a principal-agent framework a brief overview of the dimensions of information
asymmetries is given in Figure 2-3 as well as an outline of different subsets of the dissertation’s general research object.
Neo-institutional economics attempt to explain the evolution and existence of institutions (e.g. companies or market places) by mapping interdependences between these
institutions to human behavioral characteristic. It primarily consists of the transaction
costs approach, the property rights theory as well as the principal-agent theory
(Müller-Stewens and Lechner, 2003, pp. 149). Based on the works of Coase (1937)
and Williamson (1975; 1985) the transaction costs approach explains economic exchanges between two parties through the transaction costs incurred. The basic assump-
Research Set Up
15
tion in this context is that transaction costs are influenced by various factors relating to
either transaction specific or involved party specific characteristics (Erlei and Jost,
2001, pp. 36). The property rights theory is a second pillar of the neo-institutional economics perspective and is grounded on the idea that all goods can be viewed as property rights. Thus, the exchange of goods becomes an exchange of property rights. These
property rights in turn are key factors to determine the value of a good (Bernet, 2003,
p. 97).
The principal-agent theory is build upon the relationship between a principal and an
agent, whereas information asymmetries as well as conflicts of interest arise. The most
notable proponents of a principal-agent theory in the financial literature were Jensen
and Meckling (1976), Fama (1980) and Fama and Jensen (1983). In the course of this
thesis proposal, the principal-agent theory is used as a theoretical framework and applied to secondary credit markets accordingly. By definition the principal and the
agent only in part share the same objectives. Different objectives in turn lead to conflicts of interests between the two parties (Trezzini, 2005, p. 51). Conflicts of interest
in turn are impacted by information asymmetries. Existing uneven information distribution eventually allows the agent to pursue – knowingly or unknowingly by the principal – his own objectives. Basically, the principal-agent theory thus relies on the assumption that a certain degree of information asymmetry exists between the principal
and the agent (e.g. Jensen and Meckling, 1976; Fama, 1980).
The triangle at the top of Figure 2-3 illustrates a basic relationship set-up at the level of
secondary credit markets in general and consists of an information agent, an investor
and a borrower. Each party is linked to the others and thus provides us with three different principal-agent relationships. Subsets I & II directly correspond to the CDO and
CDS markets, which were already defined in chapter 2.2 as the dissertation’s core research topics. Subset I is applied to the guiding research questions (I) and (II), whereas
Subset II is applied to the third one. Depending on the underlying research focus, the
composition of the observed relationship triangle may vary. Since an investor or an
issuer appears as an involved party in both market segments, both are present in all
three relationship triangles. In Subset I the rating agency assumes the part of an information agent. Given the research focus of chapter 6 (e.g. spill-over effects between
16
Research Set Up
Figure 2-3: General Research Object and the Corresponding Subsets
General Research Object
(Information Asymmetries on Secondary Credit Markets)
Information Agent
Principal-Agent
Relationship
Issuer
Investor
Subset I
Subset II
(Collateralized Debt Obligations)
(Credit Default Swaps)
Rating Agency
Stock Analyst
Principal-Agent
Relationship
Principal-Agent
Relationship
Issuer
Investor
Issuer
Investor
Source: own illustration.
CDS spreads and stock analysts’ forecasts) this role is performed by a stock analyst in
Subset II.
Starting with Subset I (CDO market) the basic principal-agent relationship exists between the investor (principal) and the issuer (agent). The investor buys a specific
tranche in the course of a CDO transaction which in turn is issued by the investor. The
second principal-agent relationship exists between the investor (principal) and the rating agency, with the investor delegating monitoring activities to the rating agency (e.g.
Partnoy, 2006; Mason and Rosner, 2007; Güttler, 2008). A third relationship focuses
on the interaction between the issuer (principal) and the rating agency (agent). Since a
Research Set Up
17
specific characteristic of the CDO rating market is the fact that the issuer pays the rating agency and thus directly controls the interaction with the rating agency, the issuer
acts as a principal in this context (e.g. Morkötter and Westerfeld, 2008).
In Subset II (CDS market) the basic relationship between the investor (principal) and
the issuer (agent) can also be detected. Throughout a CDS contract (see chapter 2.2)
the protection seller agrees to bear a certain credit risk in exchange for an annual payment (e.g. Hull, 2006, p. 746). Since the protection seller puts a certain amount of his
capital at risk and in exchange receives a certain premium, the protection seller in turn
is acting as an investor, whereas the protection buyer acts as an issuer underwriting
credit risk. Both issuer (protection buyer) and investor (protection seller) maintain a
principal agent relationship with the information agent. Since information asymmetries
are analyzed between CDS markets and stock analysts’ forecasts, the role of an information agent is performed by a stock analyst. Again, the relationship between the investor and the information agent can be described with the investor acting as a principal, delegating monitoring activities to the stock analyst (agent). This relationship also
holds for the interaction between the issuer (principal) and the stock analyst (agent). If
the issuer is a bank and the stock analyst works for the very same institution, the principal-agent relationship is of particular interest. To overcome potential conflicts of interest arising out of this constellation, Chinese walls are implemented within the banks
concerned. With the inclusion of a stock analyst acting as an information agent, we extend the perspective towards the stock market and leave the intra-market perspective
that is limited to secondary credit markets only. Since the dissertation’s aim in this
context is to analyze spill-over effects between stock analysts’ forecasts and CDS
spreads, the main research focus (information asymmetries on secondary credit markets) is viewed to remain unchanged.
In both subsets the illustrated principal-agent relationships are used in order to analyze
the existence of information asymmetries on CDO and CDS markets. Thus, corresponding market functionality and structure as well as allocation of resources and cost
issues are addressed accordingly. Each of the outlined subgroups will be analyzed,
tested and illustrated through a specific empirical setting.
18
Research Set Up
2.4 Scientific Objective
The overall scientific objective relates to the analysis of how information asymmetries have an impact on the structure, functionality and pricing patterns of CDO and
CDS markets as well as the behavior of relevant market participants. This general
scientific objective is concretized along a descriptive, analytical and explanatory
dimension.
The scientific objective finally can be summarized under the guidance of the following
scientific statements corresponding to the thesis’s framework defined earlier (Bernet,
1982, p. 16):
(I)
descriptive scientific objectives
(II)
analytical scientific objectives
(III)
explanatory scientific objectives
The broad scientific objective relates to the question whether information asymmetries
exist and how these information asymmetries impact the structure, functionality and
pricing of CDO and CDS markets as well as the behavior of relevant market participants. Descriptive scientific objectives focus on specific characteristics of the secondary credit markets as well as its functionality and structure. The analytical objective in
turn is devoted to the analysis of different elements (e.g. market players) and their interaction with the dimension in which the research topic is integrated. While the analytical objective mainly relies on the design of the secondary credit market system, the
explanatory objective deals with interactions, system-imbedded reactions and behavioral patterns regarding secondary credit markets. In the following each of the objective dimensions is briefly reviewed against the background of the individual empirical
analysis performed in chapters 4 to 6.
The first guiding research question primarily focuses on the impact of information
asymmetries on CDO markets. It is analyzed how information asymmetries determine
market functionality and the behavior of market participants in the case of CDOs (de-
Research Set Up
19
scriptive objective) throughout rating model arbitrage. Against the background of an
analytical objective this research proposal aims to shed light on the interaction of the
triangle elements in Subset I (see Figure 2-3) and their link to the economic environment. Relating to an explanatory objective, the paper examines how the given structure
of the CDO rating market induces a specific behavior on the part of the involved parties.
Throughout the analysis of multiple CDO ratings the second guiding research question, in chapter 5, evaluates the impact of information asymmetries on the pricing
structure of underlying CDO tranche spreads. Besides the specification of multiple
CDO ratings (descriptive objective) the research proposal also investigates parts of the
relationship triangle of Subset I as well as its interacting elements (analytical objective). From the explanatory perspective the empirical analysis of chapter 5 intends to
examine how the predefined interacting elements ultimately impact the pricing structure of CDO transactions.
Finally, the third guiding research question focuses on spill-over effects (co-movement
and lead-lag structures) between stock markets (represented by stock analysts) and
secondary credit markets (represented by the CDS market). The general scientific objective relies on the question to what extend information asymmetries between secondary credit markets and stock markets exist and whether this misaligned distribution
of information leads to specific adjustment processes (lead-lag structures) in CDS
spreads. The descriptive objective therefore relies on detecting a conjunction between
these two elements of two different financial markets. The analytical perspective is
represented in the question whether information distribution is bounded by different
asset classes or whether stock analysts also perform the role of information agents in
the case of CDS spreads. Finally, the paper considers whether information asymmetries are a reliable framework for explaining the detected spill-over effects between the
different elements and thus represent an explanatory objective.
20
Information Asymmetries on Secondary Credit Markets
3 Information Asymmetries on Secondary Credit Markets
3.1 Information Asymmetries
3.1.1 Basics
Information asymmetries arise if the information distribution between two (contract)
parties is not equal. One party possesses information, which is not accessible and thus
unknown to the counterparty (Trezzini, 2005, p. 50). Information asymmetries can be
used as a competitive advantage by the party which benefits from an unequal distribution of information. If not induced otherwise the party possessing additional information has no incentive to level the existing information asymmetry (Pictot and Meier,
1993, pp. 31).
As one of the first academics Akerlof (1970) introduced the concept of information
asymmetries to an economic setting. His “lemon market” example analyzed the US
automobile market and illustrated the issue of uncertainty between two contract parties
(e.g. buyer and seller of a car). In relation to the contract formation information
asymmetries may arise ex-ante, ex-interim or ex-post (Bader, 1986, pp. 22). By definition information asymmetries can also arise between market participants on financial
markets in general and on secondary credit markets more specifically (Gerster, 2005,
p. 132). The key issue in this context and also one of the thesis’s research guidelines is
the question if the better-informed market participant capitalizes on his information
advantages and thus influences market structure and pricing patterns. Against the
background of this research question a new strand of financial literature has emerged
during the last 30 years aiming at financial asymmetries on capital markets and credit
markets in particular (e.g. Leland and Pyle, 1977; Flannery, 1986; Fazzari and Athey,
1987; Umlauf, 1991; Goswami et al., 1995; Chae, 2005; Sufi, 2007). Of course information asymmetries can also be applied to the area of risk transfer and securitization
(e.g. Franke and Krahnen, 2008). In line with Arrow (1985, pp. 38) and Spremann
(1990, pp. 563) it is distinguished between the following three types of information
asymmetries: quality uncertainty, moral hazard and hold-up.
Information Asymmetries on Secondary Credit Markets
21
The main issue in order to overcome existing information asymmetries is the trustworthy transfer of information. In principle three different concepts can indentified in order to overcome information asymmetries: signaling, delegated monitoring and reputation.
If the costs triggered by incorrect information are higher than the anticipated (financial) advantages through the incomplete information, signaling instruments may be an
appropriate measure to guarantee a credible information transfer. Thus, companies
with a bad credit quality cannot imitate the signals coming from companies with a high
credit quality (Heinke, 1998, p. 202). In case of credit markets, signaling approaches
can be differentiate as follows: If the borrower (agent) sends the signals by himself to
the investor (principal) we define it as direct signaling. Are the signals reaching the
signals from a third party we label it indirect signaling (Hsueh, 1986, p. 33). Against
the background of the defined research object both rating agencies and stock analysts
may act as third parties.
The idea underlying the concept of delegated monitoring is that the principal refers to
a third party in order to monitor the agent’s activities. Thus the principal outsources
his due diligence commitments. Specifically, in case of one agent (issuer) but several
principals (investors) the advantages of delegated monitoring are rather obvious because monitoring costs only arise once if the principals decide to assign only one third
party to perform monitoring activities. However, critical issues are the trustiness of the
third party as well as a free-rider problem (Aulibauer and Thiessen, 2002, pp.23).
A third approach in order to reduce information asymmetries is reputation of both the
agent and the involved third party performing delegated monitoring activities. Basically key is that missing transparency is overcome by the counterpart’s reputation or trustiness respectively. Reputation has typically been build up by a longstanding track
record. In case of CDO markets the issuer’s reputation is for example closely connected to the number of underdone transactions or the backing of a renowned bank.
The existence of information asymmetries requires interacting relationships between
two parties. As previously outlined in chapter 2.3 secondary credit markets offer a variety of different relationship settings and have been aligned to a principal-agent
framework. Based on these considerations throughout chapter 3.3 the detected princip-
22
Information Asymmetries on Secondary Credit Markets
al-agent relationships on secondary credit markets are addressed in more detail from
the perspective of information asymmetries. By doing so, chapter 3.1 will first briefly
review the basic theoretical principles of information asymmetries followed by an introduction of the information agents which are involved according to the dissertation’s
research object.
3.1.2 Quality Uncertainty
Quality uncertainty refers to ex-ante information asymmetry. Prior to the contract formation the contract partners are unsure about the performance quality of the respective
counterpart (Schiefer, 2008, p. 36). Besides Akerlof (1970) most notably Leland and
Peyle (1977) and Ramakrishnan and Takor (1984) introduced the concept of quality
uncertainty to financial literature. Based on theoretical models they prove that quality
uncertainty is present on financial markets and directly impacts the behavior of market
participants. The model of Ramakrishnan and Takor (1984) assumes that in this context information brokers just like rating agencies provide valuable services on the financial markets. By signaling attributes information brokers are able to reduce the level of quality uncertainty. Quality uncertainty in turn may lead to adverse selection,
representing a situation in which supply of high quality performance or products deteriorates due to comparably low pricing levels (Kiener, 1990, pp. 24). According to
Spremann (1990) quality uncertainty can also be denoted as hidden characteristics.
3.1.3 Moral Hazard
Moral Hazard describes the uncertainty of the contract partner’s behavior from an expost perspective. Thus, moral hazard is a risk parameter which becomes important after the (financial) contract is signed between the two parties (Trezzini, 2005, p. 56).
Ex-post one contract party is able to see and evaluate the outcome (e.g. return on an
investment) but not the action performed by the counterparty to achieve this outcome.
In addition, one contract party cannot verify if the outcome is linked to the actions carried out by his contractual counterpart or if the outcome is merely the result of external
impact factors which are beyond the contract partner’s influence (Mirrlees, 1999). Arrow (1985) differentiates between two kinds of moral hazard: hidden information and
hidden action. In the case of hidden information one contract party does not disclose
Information Asymmetries on Secondary Credit Markets
23
the full range of his options and the corresponding risk factors. Hidden action in turn
refers to the problematic situation that one contract party can choose options that are
not in the interest of the counterparty but are beyond his observation focus. Against the
background of a borrower-lender relationship, moral hazard may arise through actions
undertaken by the borrower, which are not observable and thus not manageable by the
lender (Freixas and Rochet, 1997, pp. 108). Many scholars have analyzed the impact
of moral hazard on economic environments. In this context, one field of financial literature focuses on credit markets and analyzes the lender-borrower relationship in more
detail (e.g. Diamond, 1984; Breuer, 1995; Gorton and Pennachi, 1995; von Thadden,
1995).
3.1.4 Hold-Up
Hold-up refers to the ex-interim perspective (e.g. maturity of loan). Both parties have
signed a (loan) contract. Since there are limitations on how precise a contract can be in
defining every detail of the contract partner’s performance and actions, there is room
for interpretation (Hartmann-Wendels et al., 2007, pp. 99). In this context, hold-up is
defined as the behavior of one contract partner to capitalize on contract gaps. The disadvantaged contract partner in turn has no possibility to react accordingly (Bernet,
2003, p. 92). Typically, the disadvantaged contract partner is the party with the greatest exposure in relation to the contract and has made significant investments or has allocated capital. If this is the case, these investments can be regarded as sunk costs
(Signer, 2003, p. 125). Against the background of a borrower-lender relationship it is
primarily the lender, who suffers from Hold-Up, since he has to put up capital.
3.2 Information Agents on Secondary Credit Markets
3.2.1 Credit Rating Agencies
Credit rating agencies (CRAs) analyze and assess the creditworthiness and the financial obligations of entities (Frost, 2006). According to Ashcraft and Schuermann
(2008) a credit rating represents an aggregated opinion of the analyzed creditworthiness reflecting only credit or default risk. Against the background of structured resale
activities, it important to note that the rating always reflects the creditworthiness of the
24
Information Asymmetries on Secondary Credit Markets
analyzed debt instruments (e.g. CDO tranche) and not of the overall credit worthiness
of the SPV. Typically, credit ratings are ordinal measures and each rating agency uses
a rating scale in order to aggregate their credit assessment into a comparative measure
(e.g. AAA). The rating scale allows investors to put the individual rating of the financial obligation into an overall context (Metz, 2007). Through the publication of credit
ratings, CRAs are mandated to create transparency on financial markets, lower existing information asymmetries between issuers and investors and, finally, improve asset
allocation on a macroeconomic level (Behrends, 1998, p. 158). From an investor’s
perspective CRAs perform an information function as well as a monitoring function.
The CRA collects, aggregates and assesses data and typically provides all investors
with this information. It would be rather costly for investors to assess the creditworthiness of entities and their financial obligations on their own. Therefore investors have a
rational reason to outsource this task with the CRA ultimately becoming a screening
instrument (Heinke, 1998, pp. 36; Ahscraft and Schuermann, 2008). With the introduction of Basel II the role of CRAs became even more important because regulatory capital requirements of financial institutions are now linked to credit ratings (Basel Committee on Banking Supervision, 2001; Perraudin and Taylor, 2004).
It is worth mentioning that the CDO rating market can be described as oligopolistic,
with only three suppliers providing rating services: Fitch, Moody’s and S&P (Partnoy,
2006; Mason and Rosner, 2008). The process leading towards the final rating is denoted in the following as a rating process. All three rating agencies maintain a sophisticated rating process with each of them widely accepted by investors. Thus, each rating agency is a viable choice for the investors in terms of due diligence delegation.
Two of the three rating agencies (Fitch and S&P) rely on an expected loss (EL) based
approach throughout the rating process whereas Moody’s rating methodology is centered on a probability of default (PD) concept. The whole process is characterized by
close exchange and interaction between the rating agency, the issuer as well as other
involved parties like asset managers or other (investment) banks. Typically, the CDO
rating process starts with the assignment of the rating agency by the issuer. At this
point it is emphasized that CDO ratings are solicited evaluations that are initiated by
the issuer. In addition, the issuer also decides whether a rating is published or not once
Information Asymmetries on Secondary Credit Markets
25
the rating is finalized by the rating agency. Due to ongoing developments on CDO
markets the applied rating models are continuously developed further (Morkötter and
Westerfeld, 2007). For a detailed description of the different rating processes as well
as underlying models are referred at this point to the three rating agencies, (Fitch,
2009; Moody’s 2009; S&P, 2009).
3.2.2 Stock Analysts
Stock analysts are the second class of information agents considered in relation to the
research object of this dissertation (see chapter 2.3). Throughout chapter 6 stock analysts assume the role of information agents on CDS markets, allowing us to observe
the interaction between CDS spreads and the assessment of the underling reference
entities by stock analysts through their earnings forecasts. Typically, investors do not
have the time nor the resources to analyze all stocks independently and thus delegate
monitoring activities to stock analysts – similar to the case of rating agencies. Like ratings, stock analyst reports are thus perceived to incorporate signaling attributes with
analysts functioning as opinion leaders on stock markets (Beier-Middelschulte 2004, p.
76).
In contrast to rating agencies stock analysts do not focus on the likelihood of a company repaying its outstanding debt but on the (future) value of its equity or its share
price, respectively. By definition not the PD or the EL are key measures to assess the
equity value but specific accounting measures which are forecasted by stock analysts
in order to give an assessment of future stock price development. In this context one of
the most prominent as well as the most frequently occurring measures are earnings-per
share (EPS) forecasts. A common valuation tool to come up with a forecast is the discounted cash flow approach (Stickney et al., 2007, pp. 855). These forecasts additionally rely on a qualitative assessment of a company's situation as well as an industry
and economy outlook in general. Typically, stock analysts aggregate their findings or
forecasts within a stock report made available to third parties. In general, the stock report also contains the analyst’s final recommendation either to buy, hold or sell the
analyzed stock (Warburg, 2006; Balboa and Gómez-Sala, 2008).
26
Information Asymmetries on Secondary Credit Markets
Usually, stock analysts are employed by (investment) banks, which in turn use the
subsequent analyst reports as a kind of marketing instruments to promote their sales
activities (e.g. brokerage, asset management or private/retail banking). Thus, it is
usually the case that a blue chip company (e.g. Nestle) is covered by 50 or more different stock analysts. Since analysts generally publish a comparable set of estimation
figures, it is also possible to compare analysts among each other and to back test the
reliability of their forecasts (e.g. Amir and Ganzach, 1998; Bolliger, 2004).
Regulatory authorities demand that stock analysts are strictly separated from the
bank’s credit analysts through Chinese walls. By separating stock and credit analysts,
regulatory authorities guarantee that stock analysts rely on publicly accessible information only, whereas credit analysts gain access to a great deal of private information
in the course of the rating process. In this context, access to information is only one
area of potential conflicts of interest. Banks also generate significant income streams
through corporate banking activities conducted directly with the companies analyzed
by the bank’s stock analysts (e.g. lending or M&A consulting). This business relationship leads to a second area of conflict of interest since the banks’ corporate customers
might demand favorable stock reports in return for banking business. Again, through
the implementation of Chinese walls stock analysts are supposed to be protected from
coming under pressure from the bank’s corporate relationship managers demanding
favorable stock reports (Baik and Park, 2003; Bradshaw et al., 2006; Chan et al.,
2007). As a result of these potential conflicts of interest several independent companies have recently emerged offering stock analyst reports as third party providers
without an affiliation to any financial institutions and thus imitating, to a certain degree, the role of rating agencies (Cohen, 2008).
Information Asymmetries on Secondary Credit Markets
27
3.3 Information Asymmetries on Secondary Credit Markets
The theoretical concept, as developed throughout Chapter 3.1, is now applied to the
specific characteristics of information asymmetries on secondary credit markets. Based
on the two defined research subsets (see chapter 2.3), the different dimensions of information asymmetries arising within the relationship triangle of investor, issuer and
information agent are illustrated.
3.3.1 Issuer and Investor
3.3.1.1 Research Subset I
The very basic principal-agent relationship takes place between the investor (principal)
and the issuer (agent) of a CDO transaction. By definition, the issuer possesses a full
set of information with regard to the CDO transaction, whereas the investor has only
access to information, which was previously revealed to him by the information agent.
The information agent, in turn, also follows his own interests. Against the background
of quality uncertainty, the investor is thus faced with two problems: first, is the set of
information he has access to a complete one? Second, is the information provided to
him by the issuer reliable? As we will see throughout chapter 4 in more detail, the issuer tries to overcome these problems by publishing a detailed set of information prior
to the transaction. In this context, he assigns one or more rating agencies as external
parties in order to assess the transaction (signaling). However, full transparency with
regard to quality uncertainty (specifically relating to the second question) can only be
achieved if the issuer agrees to disclose his information as well as his interaction with
the involved information agents (rating agencies) completely. Moral hazard and holdup both focus on an ex-post perspective (e.g. Does the CDO manager follow the investment strategy as indicated in the presale report? Are interest payments made accordingly to the defined waterfall structure?). Both moral hazard and hold-up can be
limited in large part through the application of contractual provisions and the inclusion
of third parties, such as trustees (Morkötter and Westerfeld, 2008).
28
Information Asymmetries on Secondary Credit Markets
3.3.1.2 Research Subset II
Throughout a CDS contract, the protection seller is assumed to act as an investor and
thus performs the function of a principal, whereas the protection buyer (issuer) is
viewed as the agent. The protection seller absorbs the credit risk of a specific reference
entity from the protection seller. In this context, quality uncertainty arises with regard
to the underlying reference entity. Since Research Subset II focuses on the interaction
between the CDS markets and stock analysts, the latter are viewed as acting as information agents in order to overcome quality uncertainty. Of course, if both parties possess the identical set of information, quality uncertainty with regard to the reference
entity affects both parties equally. If the protection seller only relies on stock analysts
or other publically available information sources (e.g. rating agencies), whereas the
protection buyer has access to private information, the issue of quality uncertainty becomes even more complicated. This problem setting typically appears: if the protection
buyer is a bank buying protection for a loan, the bank has originally lent to a corporate
client. Based on the detailed research gathered throughout the lending process, the
bank has access to a more detailed set of information as the protection seller. Besides
quality uncertainty, moral hazard and hold-up also become eminent. Once the CDS
contract has been signed, the protection buyer is ensured against the underlying credit
risk and thus has relatively fewer incentives to monitor the underlying credit facility
(loan) furthermore as orderly as the protection buyer has done in the past, when he
bore the credit risk all by himself (e.g. control of proper interest payments). Typically,
large parts of quality uncertainty and moral hazard are overcome by establishing Chinese walls within the protection buyer’s organization (e.g. the loan’s credit officer
does not get to know that protection was bought for a specific facility) as well as the
guarantee to perform a certain set of services throughout the CDS contract’s maturity
(e.g. Westerfeld, 2004, pp. 101). Hold-up, in turn, is experienced by the protection
buyer. Given the structure of a CDS contract, the protection buyer pays the protection
seller each year a certain premium, but he has no guarantee that the protection seller besides his contractual obligations - is able to cover the incurred loss in case of a default event (e.g. counterparty risk). Arising hold-up can primarily be overcome by the
renowned reputation of the protection buyer (e.g. high solvency).
Information Asymmetries on Secondary Credit Markets
29
3.3.2 Investor and Information Agent
3.3.2.1 Research Subset I
It is market standard that CDO investors delegate due diligence activities in large parts
to the rating agencies. Investors again are faced with information asymmetries relating
to quality uncertainty: are the ratings reliable? Due to their primarily contractual relationship with the issuer, do the rating agencies face incentives to adjust their rating
outcomes? Moral hazard is related to the fact that the investor only sees the rating outcome (e.g. AAA) but is not involved in the process leading to the final rating. The rating agencies repeatedly address the issue of information asymmetries by firm internal
corporate governance practices and by making the rating processes in large part publicly accessible. In addition, the rating agencies repeatedly argue that their neutrality is
key to their business model. Violating neutrality would, in turn, damage their overall
reputation as information agents. In the course of the current financial crisis, both the
issuer-pays-model as well as the concept of self-regulation has come under increased
scrutiny. Since the applied rating models did not capture the full dimension of the underlying risk structures, the investors were very recently affected by quality uncertainty, as the provided information (e.g. ratings) proved to be less reliable than indicated.
3.3.2.2 Research Subset II
Throughout Research Subset II, the role of the information agent is performed by stock
analysts, allowing us to link the stock markets with the CDS markets. As already observed, in the case of the CDO markets, there exists a non-contractual-based principalagent relationship between the investor and the stock analysts. Nevertheless, this induces that the investor turns to the stock analyst throughout the process of asset allocation. It is widely accepted that stock analysts incorporate guidance attributes with regard to stock markets. As discussed in Chapter 6 in more detail, these guidance
attributes are extended to the CDS markets (see Research Subset II). As also illustrated, in the case of the CDO markets for the interaction between the investor and
rating agency, the principal-agent relationship between the investor (principal) and the
stock analyst (agent) primarily includes due diligence activities performed by the stock
analyst on behalf of the investor. Relating to quality uncertainty, it is thus rather diffi-
30
Information Asymmetries on Secondary Credit Markets
cult for the investor to assess the stock analysts’ capabilities. Stock analysts can overcome quality uncertainty, for example, by referring to past forecast accuracy or specifically quoting independency. The same is true for moral hazard, which may arise, given the fact that stock analysts are, in most cases, employed by banks. Typically, these
banks are also involved in underwriting business, which might result in a potential
conflict of interests: corporate clients might ask for favorable analyst reports in return
for banking business. This kind of biased forecast would, of course, harm investors,
who would rely on misleading information throughout their allocation process. Since
Research Subset II merges with the CDS markets as well as the stock analysts’ two
sub-segments of the capital markets, it is also worth considering the interacting dynamics between these two adjacent markets. In this context, information asymmetries
may arise if one market has (early) access to information or incorporates exclusive information. Following this argumentation, we would expect one market to lead the other, whereas equal information distribution or identical access to information would
suggest a non-lagged co-movement.
3.3.3 Issuer and Information Agent
3.3.3.1 Research Subset I
As already noted by describing the principal-agent relationship between the issuer and
the investor, the rating agencies are engaged by the issuer in order to reduce the existing information asymmetries (e.g. quality uncertainty) in the direction of the investors.
Whereas rating agencies are viewed from the investors’ perspective primarily as information agents performing monitoring activities, the rating agencies are used from
the issuer’s perspective with respect to signaling attributes. In more detail, rating agencies are assigned by the issuer (principal) in order to send a strong signal with regard
to the underlying credit quality. Through the ensuing sales process prior to the issuance of CDO tranches, the issuer uses the rating outcomes as a marketing instrument
to build up trust with the investor regarding the issuance’s overall quality. With regard
to quality uncertainty, again, the principal (issuer) needs to trust the rating agency with
regard to the rating outcomes. Since CDO ratings are solicited ones, the issuer possesses in addition to the investor an additional option: the rating publication is contin-
Information Asymmetries on Secondary Credit Markets
31
gent on the issuer’s approval. Thus, if the issuer does not approve the rating outcome,
he can detain publication. Of course, this option is closely connected to the quality uncertainty as observed in the interacting dynamics between the issuer and investor.
Nevertheless, since the issuer pays the rating agency, the corresponding relationship
falls victim to serious conflicts of interest on the agency’s side: the rating agency is
getting paid by the issuer, who also decides on future rating assignments. Thus, the
dependency is rather obvious. However, at the same time, investors expect the rating
agencies to deal properly with the arising conflicts of interest and prepare an unbiased
assessment of the CDO transaction.
3.3.3.2 Research Subset II
In the case of CDS markets, the interaction between the issuer and the information
agent (represented by the stock analysts) is less pronounced as observed in the case of
the CDO markets. CDS contracts are not directly assessed by information agents (neither by rating agencies nor by stock analyst). Information agents focus on the underlying reference entity and, in the case of stock analysts’ forecasts, even on the equity
value rather than on the company-specific credit risk, as done by the rating agencies.
Secondly, opposing the business model of the CDO rating market, stock analysts’
forecasts are not solicited ones. Thus, the issuer does not pay for them nor has he an
option to block publication. This, in turn, reduces the level of potential conflicts of interest as well as information asymmetries notably. On the one hand, this makes the vital principal-agent relationship between the protection seller and the protection buyer
less pronounced to information asymmetries or conflicts of interest. However, if in
turn the protection buyer and the stock analyst’s employer are the same party (e.g. a
bank with activities both in the CDS as well as the equity market) conflicts of interest
may arise from another angle (see previous chapter).
32
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
4 Rating Model Arbitrage in CDO Markets: An Empirical
Analysis
4.1 Introduction
In this article we analyze whether information asymmetry between issuers and investors of CDO transactions leads to rating model arbitrage in CDO markets. 1 Rating
model arbitrage is based on the issuer’s option to either publicize a solicited presale
report and the underlying tranche ratings or to refrain from publication if the assigned
presale report and/ or the rating deviates from expectations. Should the issuer choose
to opt for a withholding of the results, rating model arbitrage in the applied definition
occurs as specific transactions are rated by specific agencies and/or methodologies.
We develop two hypotheses based on asymmetric information distribution between
issuers and investors to find empirical evidence for rating model arbitrage (sometimes
also referred to as “rating shopping”). Using a data sample of CDO transactions
grouped both by rating agencies and underlying rating methodologies, we test for homogeneity of characteristic transaction features in the group and heterogeneity between the different groups. Apparently, such a test has never been performed before.
We test the null hypothesis that rating model arbitrage on the basis of information
asymmetry does not exist. The alternative interprets common patterns in the characteristics of CDOs rated by the same rating agency, or with the same underlying rating
methodology as a manifestation of rating model arbitrage. We find that the null hypothesis can be rejected as individual patterns of transaction characteristics within each
group could be identified. Furthermore, we show that transactions rated by Fitch and
Standard & Poor’s (S&P) incorporate a higher degree of consistency in terms of comparable characteristics (e.g. tranche structure) than transactions rated by Moody’s. The
results support the existence of rating model arbitrage in CDO markets and deliver
useful insights for future rating commissioning and investors’ behavior.
CDOs have become increasingly important in today’s financial markets, with issuance
levels growing at an extraordinary rate in the past few years. Their development has
1
The following chapter represents joint work by Morkötter and Westerfeld (2009b).
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
33
been accompanied by debates involving market participants, regulators, politicians,
and researchers alike (specifically during the financial crisis starting in 2007/08) on the
methodologies and processes by which credit rating agencies evaluate the creditworthiness of such securities. Rating risk in the CDO context arises from the fact that the
structured nature of CDOs limits the usefulness of their ratings, since ratings only reflect certain aspects of a CDO’s credit risk properties. Ratings reflect the average risk
of a security and represent an opinion on the probability of default (PD) and expected
loss (EL). They do neither factor in the dispersion of risk around its mean, nor can they
convey the complexity of a structure or sensitivity of the structure to the embedded assumptions; for example, default correlations and recoveries post-default (see for example Chan-Lau and Ong, 2006). Furthermore, information is distributed highly
asymmetric between investors and issuers with only limited possibilities to overcome
these complexities. In cases where investors rely on ratings for their CDO investments
(usually implying a high number of underlying loans), an additional specific model
risk might arise from the respective agency’s model used to assess CDO transactions
and structures.
Moody’s Investors Service (2007a) analyzed a database of 50,000 tranches of structured finance transactions rated by both Moody’s on the one hand, and by S&P and/or
Fitch on the other, containing all ratings outstanding on February 28, 2006. The report
concluded that for jointly-rated CDOs the average rating gap vis-à-vis both S&P and
Fitch is insignificantly different from zero. Differences in ratings are greater for ratings below Aaa than for Aaa ratings, while roughly 98% of the Moody’s/S&P and
Moody’s/Fitch ratings are the same when the Moody’s rating is Aaa. The percentage
of identical ratings drops to 60% and 55% vis-à-vis S&P and Fitch respectively, when
Moody’s is non-Aaa. Differences, while very large in many cases, are likely to understate the given options, because rating model arbitrage often causes large differences in
rating opinions to be unobserved by the market. The reason is that rating model arbitrage in structured finance is interpreted to hide large systematic differences in rating
opinions across agencies. However, differences in ratings of CDO tranches may be
caused by a variety of reasons, e.g. methodological differences, adverse selection of
collateral by the issuers, differences in criteria among the agencies, differences in
34
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
monitoring practices, or an “unrepresentative” sample of securities rated by the respective agencies.
It is assumed that rating model arbitrage in the context of CDOs exists and becomes
evident in a sample selection bias. Cantor and Packer (1997) define a sample selection
bias in connection with credit ratings. Not all firms receive a credit rating from each
rating agency. Therefore, the published set of ratings is an incomplete sample. Accordingly, a sample selection bias arises in CDO markets since the sample of transaction
ratings and especially the sample of presale reports is not complete: Particular agencies
are chosen to rate a structure and publish a presale report. This suggests that the chosen agency’s rating may be higher than the rating that would have been assigned by
another agency. If an issuer requires only one or two ratings, but solicits proposed ratings and presale reports from multiple agencies, he has an incentive to choose the
highest rating or the most favorable presale reports respectively.
This chapter is organized as follows. Based on a literature review in chapter 4.2, we
develop two competing hypotheses based on information asymmetries within CDO
markets in chapter 4.3. The data sample, empirical results and interpretation of the results are contained in chapters 4.4 and 4.5. Chapter 4.6 concludes the paper.
4.2 Literature Review
CDO rating methodologies applied by the major three rating agencies differ substantially, which can result in clear differences in the ratings assigned by the agencies to
certain tranche structures (see for example Peretyatkin and Perraudin, 2002). Moody’s
has long relied on an EL criterion, as opposed to a criterion that focuses primarily on
PD, as applied by its competitors S&P and Fitch. This, however, implies that senior
instruments typically having thick tranche sizes not only show low probabilities of
loss, but generally also suffer smaller proportionate losses in the event of default. (This
does, albeit, depend partly on the transaction type: A small loss given default (LGD)
would be more characteristic of structures backed by a well-diversified pool.). Other
things being equal, an EL approach may therefore be more favorable to large senior
tranches than a default probability approach, and less favorable towards more junior
tranches that tend to be of thinner size. Fender and Kiff (2005) explore the impact of
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
35
differences in methodologies across rating agencies for senior tranche rating outcomes.
They conclude that because investors do not fully understand the possible implications
of the effects analyzed for tranche ratings, rating model arbitrage is a theoretical possibility. In this context, rating model arbitrage arises, since investors do not distinguish
between the methodologies based on EL and PD. Issuers would have an incentive to
minimize their funding costs by tailoring deal structure and strategically selecting rating agencies to obtain favorable ratings on particular tranches, due to differences in
modeling pooled credit risk. In practice, however, the authors could only find limited
evidence for this behavior.
In a letter to the Securities and Exchange Commission Moody’s Investors Service
(2007b) comments on proposed rules regarding control of credit rating agencies. It
terms rating model arbitrage as issuers’ behavior prior to the transaction’s finalization
by asking different rating agencies for the possible rating outcome, but only requesting
ratings from selected rating agencies. Moody’s proves the existence of the defined rating model arbitrage habit by identifying 44 residential mortgage-backed securitizations, in which Moody’s ratings were not accepted but, had they been chosen, would
have resulted in significantly lower outcomes than the ratings of the actual chosen
agencies.
In contrast to Fender and Kiff (2005), we empirically test for the existence of rating
model arbitrage in the context of CDOs. In addition to the above-mentioned study undertaken by Moody’s Investors Service (2007b), this is the first empirical work to analyze the existence of rating model arbitrage based on an extensive data pool and analyzing the specific patterns of the transactions’ characteristics.
Looking closer into finance literature, various papers apply the theory of asymmetric
information based on Jensen and Meckling (1976) in a credit and financing context.
Flannery (1986) develops a theoretical concept to show the relationship between
asymmetric information and risky debt maturity choice. He contends that if firm insiders are systematically better informed than outside investors, they will choose to issue
those types of securities that the market appears to overvalue most. Knowing this, rational investors will try to infer the insiders’ information from the firm’s financial
structure. With positive transaction costs, high-quality firms can sometimes effectively
36
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
signal their true quality to the market. The existence of a signaling equilibrium is
shown to depend on the distribution of a firms’ quality and the magnitude of underwriting costs for corporate debt (an originator’s choice of rating agency may signal inside information on the quality of the underlying pool). Goswami et al. (1995) examine the impact of information asymmetries on the design of debt contracts, with a
view to explaining features of debt financing. They show that, depending on the
asymmetry of information concentrating around long-term or short-term cash flows,
firms finance with coupon bearing long-term debt that either partially or does not restrict dividend payments. If asymmetry of information is uniformly distributed across
dates, firms tend to finance with short-term debt.
Other studies analyze the comparison of jointly-rated credit instruments. Generally,
three reasons why investors gain additional rating for bonds are discussed. First, an
additional rating may convey any incremental information to the markets that reduces
the costs of borrowing for the firm. Several papers investigate the effect of split bond
ratings. However, these papers fail to reach a consensus on how the market prices
bonds with split ratings (see Jewell and Livingston, 1999; Flannery, 1986). Norden
and Weber (2004) analyze whether prices react after a rating event, based on the assumption that credit ratings convey new information to the market. If credit ratings are
only to reflect information that is already known by the market, prices should not react
to the rating event at all. They conclude inter alia that both the CDS and the stock
market not only anticipates rating downgrades, but also reviews for downgrade by all
three rating agencies. Second, the main rating agencies might be biased or misjudge
some bond issues. For these misjudged issues, an additional rating could provide useful information that is valued by investors. Third, issuers may hunt for rating agencies
that provide inflated ratings. If the requested rating is favorable, the issuer publicizes it
and if the requested rating is unfavorable, the rating is not released. Therefore, requesting an additional rating is similar to buying an option on a rating as it raises the costs
of borrowing. This has the effect of ensuring that lower than expected ratings from the
additional agency are rarely made public. Jewell and Livingston (1999) compare bond
ratings of Fitch to those of Moody’s and S&P in order to analyze the potential benefits
of seeking out additional ratings from a smaller rating agency (Fitch), by comparing
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
37
rating levels, rating changes, and the impact of ratings on bond yields. Inter alia, the
authors test for the hypothesis that the average observed rating from Fitch is likely to
be significantly higher than the “true” average rating from the two other agencies.
Their analysis confirmed this hypothesis.
In this context, Cantor and Packer (1995) analyze whether the reason for getting an
additional rating may be regulatory in nature. Many financial institutions have limits,
either self imposed or imposed by government regulators, on the amounts of debt they
can hold of certain ratings. As most of these regulations only require that the highest or
second highest rating be above the cutoff point, the firm’s chances of meeting the
standard increase if a third or fourth rating is obtained. Therefore, issuers could have a
strong incentive to obtain multiple ratings to reach those investors. However, the authors find no evidence that firms obtaining Fitch ratings are doing so in order to meet
rating regulation requirements. In a later paper, Cantor and Packer (1997) empirically
test for the existence of rating model arbitrage in bonds. They find evidence that third
ratings in bond markets on average assign higher ratings than the first two rating outcomes and that the policy of rating on request induces a sample selection bias. This
sample selection bias is closely linked to our definition of rating model arbitrage.
4.3 Information Asymmetries within CDO Markets and the Role of
CRAs
In the course of our paper we focus on CDO transactions as well as the involved rating
agencies. A plain vanilla CDO transaction is typically centered around a SPV. This
SPV invests in various credit-linked assets (e.g. SME loans, bonds or tranches of other
CDO transactions) and refinances its’ purchases through the issuance of notes. Capital
and interests of the notes are guaranteed through the asset pool held by the SPV, whereas the asset pool’s composition defines the CDO transaction type. The asset pool of a
simple true sale CDO structure for example may consist of SME loans with true sale
referring to the fact that legal ownership of the underlying assets is transferred to the
SPV. This is opposed by synthetic transactions where legal ownership remains at the
issuer and only the credit risk of the involved assets is transferred to the SPV via
CDSs. In this context typically an investment bank or an external asset manager is act-
38
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
ing as an issuer setting up the whole transaction structure, hiring third parties to perform additional services (e.g. trust companies, rating agencies or other investment
banks) and eventually orchestrating the tranche issuance. We do not further distinguish
between arranger and issuer and assume them to be only one party. In practice, it is not
uncommon that a subsidiary of the arranger is acting as the issuer. Since in most cases
the arranger is the controlling shareholder of the subsidiary, we believe it is appropriate to treat them as a consolidated party.
During the months and years prior to the worldwide financial crisis starting in 2007/08
specifically structured CDOs became popular among issuers and investors. Since the
underlying asset pool of a structured CDO consists of tranches of other CDO or ABS
transactions, the valuation of these transactions is far more complex. With the observed low levels of liquidity in CDO markets during the crisis and due to the high degree of complexity it is rather difficult to obtain market prices for this kind of CDO
notes.
It is common market practice that the issuance of a CDO is accompanied by the publication of a rating. Ratings establish a form of due diligence delegation and provide
guidance to investors in the course of capital allocation. This general assessment of
ratings also holds for the CDO market. However, we have to take into account some
specific features which differentiate the CDO rating market from traditional bond rating practices:
First, the CDO rating market is an oligopolistic market, since only three different suppliers provide rating services: Fitch, Moody’s and S&P. Each of the three rating agencies maintains a sophisticated rating process with each of them being widely accepted
by investors. Thus, each rating agency is a viable choice for investors in terms of due
diligence delegation. Two of the three rating agencies (Fitch and S&P) rely on an ELbased approach throughout the rating process whereas Moody’s rating methodology is
centered around a PD concept.
However, the rating models used by rating agencies heavily rely on assumptions, e.g.
with regard to recovery rates and model specific assumptions such as mean reversion.
Recovery rate assumptions usually depend on the asset type, which entails the nature
of the asset (e.g. corporate) and the seniority in the issuer’s capital structure (e.g. se-
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
39
nior unsecured), and second a country parameter. While many models use a beta distribution or simply assume fixed recovery rates independent of default rates, sophisticated rating models, e.g. the model used by Fitch, even consider the correlation between recovery rates and the level of average default rates. In fact, recovery rates do
not only depend on debtor’s characteristics but also on prevailing market and economic conditions. Specifically, recovery rates tend to decline with an increasing number of
defaults. To reflect empirical data on this issue (Fitch Derivatives, 2006), some models
first determine a default distribution for the underlying portfolio and then link a conditional recovery rate, depending on the specific default rate of the portfolio, i.e. tiered
recovery rate assumption for increased stress scenarios. Even though this sophistication causes the loss distribution to have a longer tail, it must be doubted that rating
models even with tiered recovery rates are able to fully capture actual loss rates in
market turmoil.
Regarding rating model specific assumptions it needs to be mentioned that besides a
mean reversion assumption, that might fail during times of crisis especially in shortterm valuations, one crucial assumption of the rating models is that they all rely on a
structural model approach. Although this theoretical model is used in many practical
applications, some of its assumptions and restrictions might be critical. First the assumption of normality for asset returns, second the implied calculation of asset correlation estimated by equity correlations and third the default point, below which the
firm is supposed to default, which is difficult to be calculated accurately. In particular,
the correlation assumptions should be exposed to stress scenarios as these parameters
heavily influence the models and vary considerably through different business cycles.
Second, besides the rating models used, CDO ratings are solicited ones initiated by the
issuer. In addition, the issuer also decides whether a rating is published or not. This
implies that actually not all rating outcomes are made public as the issuer has a strong
incentive to only publish rating outcomes in favor of the transaction. If this is not the
case the issuer might withhold ratings and benefit from this behavior by paying a lower risk premium to investors.
Third, the rating outcome can be divided into publication of a presale report on the one
hand and publication of individual tranche ratings on the other hand. The presale re-
40
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
port analyses the CDO transaction in detail including legal aspects and thereby establishes the most important source of information for potential investors. Typically, a
presale report comes along with tranche ratings of the underlying transaction. Contrarily, it is common to only publish tranche ratings but not a corresponding presale report
illustrating the transaction details. In these cases a presale report might exist but the
issuer has limited information sharing to transaction ratings as he has incentives to do
so. Thus, published tranche ratings without accompanied presale report are not a sufficient indicator for the lacking existence of a presale report. However, in this paper we
define a full CDO rating as the publication of a presale report including tranche ratings.
Fourth, CDO rating processes involve a high degree of interaction between issuer and
rating agency, thereby creating potential for conflicts of interest. The negotiation phase
either leads to a presale report including rating assignment or cessation of negotiations
when consensus cannot be found (Moody’s Investors Service, 2007b). These dialogs
are time-consuming and bear significant costs for the issuer, amounting to approximately 4.25 bps of the transaction’s par value (Standard & Poor’s, 2007). However,
issuers willing to pay a rating fee gain the benefit of a solicited rating process, which
allows them to put their best case before the agencies for the final evaluation (Cantor
and Packer, 1995). In CDO markets investors are kept out of the dialog between the
issuer and the rating agency. They are not aware of the number of initiated rating
processes, termination of negotiations or unpublished rating outcomes. From the investor’s perspective the only visible outcome of the rating process is the published presale
report and/or tranche ratings.
The transactional setup of a CDO is primarily centered around the issuer, the investor
and the rating agency. Based on the principals of information economics (e.g. Arrow,
1969; Jensen and Meckling, 1976; Leland and Pyle, 1976; Fama, 1980; Fazzari and
Athey, 1987; Goswami et. al., 1995), we identify three principal agent relationships:
The first between investor (principal) and issuer (agent) since the investor possesses
the capital and decides upon its allocation. The issuer in turn serves as an agent by offering an attractive investment opportunity. A second relationship emerges from the
interaction between the issuer (principal) and the rating agencies (agent). The issuer
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
41
pays for the ratings, the rating agencies in turn deliver the asked services (rating) and
thus behave like agents. Since the investor delegates his due diligence efforts to the
rating agency, a third principal agent relationship can be identified between the rating
agency (agent) and the investor (principal).
For our empirical section we focus on the principal agent relationship between issuer
and investor, which we assume to be the most important one for capital allocation and
market efficiency. In the following, we test if this relationship is biased by information
asymmetries. Specifically, we argue that the issuer obtains information during the rating process, which he only partially shares with investors, e.g. only favorable rating
outcomes are made public. Investors do not control the due diligence process even
though rating agencies perform it on their behalf. Control lies entirely with issuers
having economic incentives to keep it, i.e. lower risk premiums result if unfavorable
ratings are not published. Rational investors could demand a full set of information
which would imply full disclosure of all negotiations with rating agencies and its derived information throughout the rating process. However, current market standards
are different.
Accordingly, the issuer has a strong incentive to use the specific structure of the CDO
rating market to limit information sharing to favorable ratings. We define this willingness to deliberately capitalize asymmetric information distribution between issuer and
investor as rating model arbitrage and test the following hypothesis:
Hypothesis H0:
Rating model arbitrage on the basis of information asymmetry
does not exist.
Hypothesis H1:
Rating model arbitrage on the basis of information asymmetry
exists.
If rating model arbitrage exists, we should be able to find patterns that are common for
CDO transactions rated by a specific agency or by a specific rating model. Homogeneity between transactions rated by a specific rating agency and/or methodology would
allow us to reject H0. If in turn rating model arbitrage does not exist, transactions with
specific features (e.g. volume, number of tranches) should be distributed equally.
42
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
4.4 Data Sample
The analysis is based on 231 international presale reports for 202 different CDOs published between August and December 2006 by Fitch, Moody’s and S&P. We decided
to use published presale reports as the basis of our analysis since a lot of CDO transactions receive two or even three ratings for their underlying tranches. Thus, multiple
ratings are a rather difficult object of investigation to identify patterns. In turn, publication of presale reports is primarily limited to one rating agency per transaction. This
fact makes us believe that information asymmetries are centered around publication of
presale reports.
In order to test our hypotheses, we apply both univariate and multivariate tests. Presale
reports are prepared by rating agencies prior to the notes’ issuance and typically published in parallel with the (preliminary) rating. Specifically, presale reports contain the
specific characteristics of a CDO and the underlying tranches. According to the specific rating scale, each presale report contains a rating outcome for the underlying
tranche. Each presale report was downloaded from the respective website of the three
rating agencies and subsequently 15 different characteristics (e.g. volume of transaction, time to maturity) including the underlying tranche structure were analyzed. Nine
out of the 15 characteristics were obtained for each of the 231 presale reports and
therefore qualify for our test section. Including the size of the underlying tranches, our
original data pool consists of 5,544 observations. Since our data sample represents all
publicly released presale reports for the time period between August and December
2006, we consider it to be a consistent data pool. Sixty-five presale reports were provided by Fitch, 59 by Moody’s and 107 by S&P. For 28 out of the 202 transactions
more than one rating agency published a presale report.
In addition to rating agencies and rating methodologies, we use the following nine characteristics:
x transaction type
x size structure of the tranches (including equity portion)
x asset management
x involved parties
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
43
x maturity of tranches
x number of tranches
x total volume
x currency
x cash flow structure.
Each characteristic is defined in detail as follows:
The CDOs are classified along four different transaction types: Collateralized Bond
Obligations (CBO), Collateralized Loan Obligation (CLO), Synthetic CDO (S CDO)
and other transactions (e.g. Collateralized Fund Obligation). The difference between
CBO and CLO can be explained by the divergent asset pool: bonds for CBOs and
loans for CLOs. Structured CDOs belong to the transaction type CBO since investments in other CDO tranches are bond investments and represent by far the largest
group within this transaction type. Simple CDOs primarily consist of loans and thus
are allocated to the corresponding transaction type CLO. In contrast, S CDOs primarily rely on CDSs as underlying.
Since rating agencies apply two different rating methodologies (PD and EL approach)
this feature proves to be helpful for analyzing CDOs. As discussed, Fitch and S&P apply the PD approach, whereas Moody’s uses the EL approach. The variable “asset
management” refers to managed or static CDOs. In a static CDO, the asset pool remains the same during the lifetime of the transaction, whereas in a managed CDO, the
asset pool’s composition might be changed, based upon variance in market conditions.
The number of involved parties in a CDO (e.g. trustee, servicer, etc.) can vary substantially between different transactions: Since additional coordination arises with each
additional party involved, complexity of CDO structure is positively correlated with
the number of involved parties. It is also not uncommon that one and the same party is
responsible for more than one function. An investment bank, respectively its subsidiary, may act as the transaction’s issuer and functions at the same time as the counterparty for swap contracts demanded by the transactional set-up (e.g. to hedge currency
or interest rate risk). If one party performs multiple functions counterparty risk in connection with this party increases negatively affecting the transaction’s overall risk profile. Thus, a low number of involved parties does not always represent a lower degree
44
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
of complexity respectively a lower risk profile. The transaction’s risk profile can only
comprehensively assessed by analyzing the transaction’s structure in detail. The maturity of the entire CDO transaction is defined as the mean of the different underlying
tranches’ legal maturities. The listing of tranches includes the equity portion. The total
volume of the tranche is denominated in EUR. If the tranches are denominated in currencies other than EUR, values were converted on the basis of the exchange rates on
October 4, 2006. In terms of the different cash flow structures, we differentiate between pass-through and pay-through transactions. In a pass-through CDO, the cash
flow is transferred directly to the investors, whereas in a pay-through construction, the
timing of cash flows is restructured. In order to incorporate the seniority structure in
our analysis, we group the ninth characteristic “size structure of the tranches” into seven different tranche classifications, plus the equity share. We therefore classify the
tranches on the basis of the seniority structure, according to the information revealed
in the presale reports into Super Senior, Class A, Class B, Class C, Class D, Class E
notes, other more subordinated notes and equity. Each tranche as well as the equity
portion is displayed as a percentage of the entire transaction volume. Following the
basic structure of seniority, Class E notes, for example, are subordinated to Class D
notes and incorporate a lower rating than that obtained by Class D. Not all of the sample’s CDOs have termed their structure of seniority in line with the aforementioned
denomination. However, in order to compare the CDOs, we have adjusted the varying
notations using our classification system as a guideline. In the case of more than one
Class A-note (e.g. Class A1, Class A2) we have aggregated the different tranches into
one Class A-note, since these notes often incorporate the very same rating and would
otherwise dilute the CDO-specific structure of seniority.
4.5 Empirical Results
4.5.1 Univariate Tests
As a starting point, sixteen variables (eight transactions’ characteristics and eight variables relating to the size structure of tranches, including equity portion) are included
to perform univariate tests for the null and alternative hypothesis. The underlying data
set is divided into two groups: The first dataset (Set I) is grouped along three data
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
45
pools, corresponding to the three rating agencies Fitch Ratings, Moody’s and S&P. In
the second set of groups (Set II), the presale reports were separated on the basis of the
applied rating methodologies (EL vs. PD approach). The EL based sub-group therefore
composes the transactions rated by Moody’s and the PD based sub-group the transactions rated by Fitch Ratings and Standard & Poor’s.
4.5.1.1 Set I (sorting by Rating Agencies)
The first step for Set I focuses on a comparison of the mean, median and standard deviation of each group. In order to normalize standard deviation, the ratio of the group’s
Table 4-1: Comparison of Subgroups (Set I & II)
Variable
(numeric & alphanumeric)
Asset Management
Cash Flow Structure
Currency
Transaction Type
Tranche Structure
Variable
(numeric)
Maturity**
(in years)
Number of involved Parties
Number of Tranches
Volume
(in mEUR)
Managed
Static
Pay Through
Pass Through
EUR
USD
Others
CBO
CLO
S CDO
Others
Super Senior*
Class A
Class B
Class C
Class D
Class E
Others
Equity
Median
Std. dev.***
Median
Std. dev.***
Median
Std. dev.***
Median
Fitch
(in %)
66.15
33.85
81.54
18.46
49.23
49.23
1.54
23.08
49.23
18.46
9.23
10.85
59.43
7.24
5.58
3.65
2.74
4.36
6.14
Moody’s
(in %)
83.05
16.95
100.00
0.00
76.27
16.95
6.78
5.08
54.24
30.51
10.17
14.75
58.42
7.98
4.51
3.20
2.52
0.41
8.26
S&P
(in %)
87.85
12.15
99.07
0.93
34.58
64.49
0.93
16.82
63.55
12.15
7.48
3.42
69.99
7.37
5.08
4.24
2.28
1.16
6.52
EL
(in %)
83.05
16.95
100.00
0.00
76.27
16.95
6.78
5.08
54.24
30.51
10.17
14.75
58.42
7.98
4.51
3.20
2.52
0.41
8.26
PD
(in %)
77.00
23.00
90.30
9.70
41.91
56.86
1.24
19.95
56.39
15.31
8.35
6.23
66.00
7.32
5.27
4.02
2.45
2.37
6.38
Total
(in %)
79.02
20.98
93.53
6.47
53.36
43.56
3.08
14.99
55.67
20.37
8.96
8.41
64.06
7.49
5.08
3.81
2.47
1.87
6.86
Fitch
32.50
0.58
5.00
0.42
7.00
0.47
501.62
Moody’s
10.00
1.17
5.00
0.22
6.00
0.31
400.00
S&P
14.00
0.91
4.00
0.44
7.00
0.24
394.86
EL
10.00
1.17
5.00
0.24
6.00
0.31
400.00
PD
16.00
1.02
4.00
0.47
7.00
0.35
405.55
Total
14.83
1.07
5.00
0.35
7.00
0.34
402.00
Std. dev.***
1.12
5.49
2.96
5.49
2.42
3.52
* For reasons of transparency we also include the Super Senior tranche, which contains of comparable rating scales as in Class A.
** Legal maturity
*** Ratio of computed std. dev. and median
46
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
standard deviation to the group’s median was named as the decisive variable. Table 41 reveals the first signs of heterogeneity between the three rating agencies. The notes’
denomination, for example, varies clearly between the three rating agencies and the
same is true for the standard deviation of the variable “volume”.
First patterns of heterogeneity can also be detected when analyzing different transaction types. As outlined before, the group “CBO” primarily consists of structured CDOs
or CDO Squared. Therefore, in the case of Fitch CBOs are overrepresented with
23.08% in comparison to Moody’s (5.08%) and S&P (16.52%).
Following the comparison of means, medians and standard deviations, we now analyze
the presale reports focusing on the univariate separation power of each characteristic.
The tests of equality of group means for Set I in Table 4-2 are based on a one-way
ANOVA including the values for Wilks’s Lambda. The results obtained provide us
with the univariate separation power of each characteristic. If we analyze the results on
thH EDVLV RI D OHYHO RI VLJQLILFDQFH RI Į HTXDO WR WHQ RXW RI WKH VL[WHHQ IHDWXUHV
have univariate separation quality. The higher end of the structure of seniority (Super
Senior and Class A notes) and the most subordinated tranches (others) incorporate
group means, which differ significantly between the three sub-groups. In contrast,
mezzanine tranches, equity portion and number of involved parties do not differ significantly and therefore incorporate no individual separation power.
Table 4-2: Test of Equality of Group Means - Set I (ANOVA)
Asset Management
Cash Flow Structure
Currency
Maturity
Number of involved Parties
Number of Tranches
Transaction Type
Volume
Super Senior
Class A
Class B
Class C
Class D
Class E
Others
Equity
* level ŽĨƐŝŐŶŝĨŝĐĂŶĐĞ;ɲсϱйͿ
Wilks’
Lambda
0.9461
0.8783
0.9296
0.8288
0.9990
0.9296
0.9654
0.9640
0.9596
0.9540
0.9983
0.9923
0.9842
0.9961
0.9004
0.9828
F
64991
157970
86363
235484
0.1193
86359
40855
42540
48015
54933
0.1947
0.8846
18341
0.4490
126132
19975
df1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
df2
228
228
228
228
228
228
228
228
228
228
228
228
228
228
228
228
Sig.
0.0018*
0.0000*
0.0002*
0.0000*
0,8876
0.0002*
0.0181*
0.0154*
0.0091*
0.0047*
0.8232
0.4143
0.1621
0.6388
0.0000*
0.1380
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
47
Since the test of equality of group means is only one part of the univariate comparison
of groups, Kolmogorov-Smirnov-tests (KS-test) are run in order to assess the difference in the data points characteristics with respect to the sub-groups. The KS-test is
chosen due to its non-parametric and distribution-free qualities. By comparing each
rating agency with the other two, we perform three different KS-tests to get the entire
picture of this test section. The results of the KS-test session – as outlined in Table 4-3
on behalf of both Set I and Set II – provide additional insight concerning the univariate
separation qualities of the different characteristics. It is only in the case of one variable
(maturity) that the outcomes of the KS-tests show that the two datasets - and therefore
all three groups - differ significantly for all three test sessions. With regards to “currency” and “number of involved parties”, Fitch and S&P differ significantly compared
to Moody’s, but in turn do not differ significantly between each other.
Since transactions analyzed by Moody’s are dominated by Euro-denominated CDOs,
we may link the observed heterogeneity to the variable “currency”. The number of
tranches, portion of Class A and the portion of other tranches also prove to separate the
presale reports on the basis of the rating agencies at least for two constellations. If the
results of the KS-test are compared with the results of the ANOVA, it can be observed
that “maturity” is identified in both test sessions as the variable with univariate separation power. With the exception of the variable “rating methodology”, this correspondence can be confirmed for all variables with one or two matches in the KS-test. Solely the variable “equity” proves to have univariate separation qualities only for the application of the KS-test. Furthermore, the majority of characteristics sorted by rating
agency differ from each other. Only the outcomes of Fitch and S&P prove to have
some degrees of concordance. The results observed therefore document homogeneity
within the sorted groups (rating agencies) and heterogeneity between the groups. The
results can be interpreted as a first sign of rating model arbitrage in CDO markets, i.e.
rejection of the null and non-rejection of the alternative hypothesis.
Asset
Management
* level of significance (ɲс5йͿ
1.0000
-0.0340
Negative
0.2253
0.0000
Positive
Kolmogorov-Smirnov Z
0.0340
Absolute
Asymp. Sig. (2-tailed)
Most Extreme Differences
Grouping Variable: Rating Methodologies (EL vs. PD approach)
Set II (Kolmogorov-Smirnov Test)
10000
0.0000
Negative
0.2960
0.0480
Positive
Kolmogorov-Smirnov Z
0.0480
Absolute
Asymp. Sig. (2-tailed)
Most Extreme Differences
Grouping Variable: Rating Agencies (Moody's vs. Standard & Poor's)
13797
0.0000
Negative
0.0444*
0.2170
Positive
Kolmogorov-Smirnov Z
0.2170
Absolute
Asymp. Sig. (2-tailed)
Most Extreme Differences
Grouping Variable: Rating Agencies (Fitch Ratings vs. Standard & Poor's)
0.3403
-0.1690
Negative
0.9397
0.0000
Positive
Kolmogorov-Smirnov Z
0.1690
Absolute
Asset
Management
Asymp. Sig. (2-tailed)
Most Extreme Differences
Grouping Variable: Rating Agencies (Fitch Rating vs. Moody's)
Set I (Kolmogorov-Smirnov Test)
0.9633
0.5001
-0.0756
0.0000
0.0756
Cash Flow
Structure
10000
0.0576
-0.0093
0.0000
0.0093
0.1667
11145
0.0000
0.1753
0.1753
0.2425
1.0267
-0.1846
0.0000
0.1846
Cash Flow
Structure
0.0000*
2.3964
-0.3615
0.0562
0.3615
Currency
0.0000*
2.5711
-0.4169
0.0585
0.4169
0.3505
0.9317
-0.1465
0.0060
0.1465
0.0217*
1.5038
-0.2704
0.0524
0.2704
Currency
0.0000*
2.4773
-0.3738
0.0111
0.3738
Maturity
0.0000*
2.3073
-0.3741
0.0170
0.3741
0.0000*
2.7209
-0.0643
0.4279
0.4279
0.0000*
3.3019
-0.5937
0.0016
0.5937
Maturity
0.0580
1.3304
-0.0882
0.2007
0.2007
Number of
involved
Parties
0.0389*
1.4037
-0.0783
0.2276
0.2276
0.9866
0.4526
-0.0510
0.0712
0.0712
0.4353
0.8701
-0.1046
0.1565
0.1565
Number of
involved
Parties
0.1635
1.1188
-0.1688
0.0121
0.1688
Number
of
Tranches
0.3616
0.9231
-0.1497
0.0111
0.1497
0.0427*
1.3870
-0.0785
0.2181
0.2181
0.0194*
1.5226
-0.2738
0.0138
0.2738
Number
of
Tranches
0.1159
1.1933
0.0000
0.1800
0.1800
Transaction
Type
0.0687
1.2982
0.0000
0.2105
0.2105
0.9551
0.5129
-0.0625
0.0807
0.0807
0.2694
1.0005
0.0000
0.1799
0.1799
Transaction
Type
0.2267
1.0431
-0.1336
0.1574
0.1574
Volume
0.2956
0.9769
-0.1286
0.1584
0.1584
0.1523
1.1346
-0.0187
0.1784
0.1784
0.1517
1.1355
-0.2042
0.1557
0.2042
Volume
0.6363
0.7446
0.0000
0.1123
0.1123
Super
Senior
0.4479
0.8616
0.0000
0.1397
0.1397
0.6430
0.7406
0.0000
0.1165
0.1165
0.8860
0.5830
0.0000
0.1048
0.1048
Super
Senior
0.0011*
1.9372
-0.2923
0.0208
0.2923
Class A
0.0001*
2.2936
-0.3719
0.0100
0.3719
0.0068*
1.6869
-0.2653
0.0840
0.2653
0.3980
0.8962
-0.1611
0.1119
0.1611
Class A
0.1075
1.2090
-0.1559
0.1824
0.1824
Class B
0.1195
1.1869
-0.1532
0.1925
0.1925
0.6800
0.7186
-0.1130
0.0774
0.1130
0.3224
0.9542
-0.1604
0.1716
0.1716
Class B
0.0607
1.3219
-0.1994
0.1088
0.1995
Class C
0.1513
1.1361
-0.1842
0.0878
0.1842
0.6862
0.7150
-0.1124
0.0966
0.1124
0.0699
1.2950
-0.2329
0.1515
0.2329
Class C
0.1233
1.1802
-0.1781
0.0604
0.1781
Class D
0.0556
1.3383
-0.2170
0.0504
0.2170
0.2610
1.0085
-0.1586
0.0296
0.1586
0.4767
0.8425
-0.1515
0.1098
0.1515
Class D
0.6385
0.7433
-0.0465
0.1121
0.1121
Class E
0.6252
0.7512
-0.0374
0.1218
0.1218
0.8741
0.5925
-0.0881
0.0932
0.0932
0.7721
0.6627
-0.0615
0.1192
0.1192
Class E
0.0044*
1.7485
-0.0058
0.2638
0.2638
Equity
0.9836
0.4611
-0.0748
0.0000
0.0748
0.0249*
1.4812
0.0000
0.2329
0.2329
0.0107*
1.6169
-0.2907
0.0000
0.2907
Others
0.2680
1.0019
-0.1512
0.0000
0.1512
Others
0.0208*
15112
-0.0108
0.2450
0.2450
0.3648
0.9207
-0.1448
0.0730
0.1448
0.0093*
1.6386
0.0000
0.2947
0.2947
Equity
48
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
Table 4-3: Kolmogorov-Smirnov- Test (Set I & II)
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
49
4.5.1.2 Set II (sorting by Rating Methodologies)
Sorting by different rating agencies already showed first signs of common patterns between the two rating agencies Fitch Ratings and S&P. The following sorting by rating
methodology therefore seeks to confirm these patterns. The comparison of mean, median and standard deviation for each rating methodology (Table 4-1) displays comparable features as observed for Set I. The means of “currency” for the two subgroups
diverge again largely. The PD approach is more often used for CBOs respectively
Structured CDOs. However, S CDOs are in turn primarily rated by the EL approach.
In accordance with the framework of the previously-implemented test design, we start
testing equality of group means (ANOVA) for Set II. The ANOVA’s results (Table 44) provide us with an analysis of the individual ability of each variable in order to separate the two subgroups on an univariate basis. Eight of sixteen variables prove to have
univariate separation power at a level of significance below 5%. In terms of identified
separation variables, this result is slightly lower in comparison with the analysis of Set
I (eleven out of sixteen). However, we have to point out that two variables (“Class A”
and “equity”) are situated in the near range of the required level of significance of 5%.
All identified variables with univariate separation power already proved univariate separation power in the test section of Set I. In contrast to the results of Set I, no univariate separation power was assigned to the variables “asset management” and “Class A
Table 4-4: Test of Equality of Group Means - Set II (ANOVA)
Asset Management
Cash Flow Structure
Currency
Maturity
Number of involved Parties
Number of Tranches
Transaction Type
Volume
Super Senior
Class A
Class B
Class C
Class D
Class E
Others
Equity
ΎůĞǀĞůŽĨƐŝŐŶŝĨŝĐĂŶĐĞ;ɲсϱйͿ
Wilks’
Lambda
F
df1
df2
Sig.
0.9986
0.9795
0.9411
0.9231
10000
0.9609
0.9657
0.9641
0.9762
0.9835
0.9984
0.9945
0.9894
0.9999
0.9710
0.9834
0.3213
47821
143431
190672
0.0001
93242
81223
85310
55775
38391
0.3772
12653
24636
0.0214
68301
38623
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
229
229
229
229
229
229
229
229
229
229
229
229
229
229
229
229
0.5714
0.0298*
0.0002*
0.0000*
0.9913
0.0025*
0.0048*
0.0038*
0.0190*
0.0513
0.5397
0.2618
0.1179
0.8839
0.0096*
0.0506
50
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
notes”. According to the performed ANOVAs of Set I and II, the number of involved
parties plays no significant role in both datasets for separating the underlying subgroups.
In addition to the ANOVA, we also perform the KS-test for Set II and determine
whether or not the selected characteristics account for the separation of the two subgroups. Table 4-3 also summarizes the results of the performed KS-test for Set II. Five
of the sixteen analyzed variables lead to univariate differentiation of the two subgroups. In comparison with Set I’s results of the KS-test, only the number of involved
parties is excluded from the selection of characteristics. Compared with the results of
the ANOVA, the variables “currency” and “maturity” separate in both test architectures the subgroups on a significant level. These findings are similar to the test results
of Set II. Also, when sorting for rating methodologies, additional support for a rejection of the null hypothesis is found.
4.5.2 Multivariate Tests
The results of the univariate test section verify homogeneity within the analyzed subgroups and heterogeneity between the different subgroups. Additionally, multivariate
tests are applied in order to identify individual patterns in the characteristics of CDO
transactions within each group. Common group-wide patterns in the characteristics are
seen as evidence for rating model arbitrage, since this implies that specific CDO transactions are rated by the same rating agency or with the same rating methodology. Using the results of the univariate tests, we pick certain variables and perform a number
of different discriminant analyses with the chosen variables. The aim is to identify the
combination of factors delivering the best subgroups’ classification of presale reports
based on discriminant analyses. This optimization problem was not approached by performing all discriminant analyses that would be theoretically possible, but by presorting the variables relying both on the results of the univariate tests and on economic
reasoning. By applying a classification based on Fisher’s linear discriminant function,
the performance of each discriminant analysis in connection with its separation power
is tested.
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
51
4.5.2.1 Set I (sorting by Rating Agencies)
The results of the different discriminant analyses for the grouping variable “rating
agencies” are shown in Table 4-5. The composition of discriminant function coefficients of each analysis can be derived from Table 4-6. For all performed discriminant
analyses, the significance on the basis of Wilks-Lambda is equal to zero and therefore
highly significant with regard to its separation qualities.
The variables of Discriminant Analysis I solely chose on the basis of the outcomes of
Set I’s ANOVA. All variables that possess a level of significance below 5% are included. Only relying on the ANOVA’s outcome already leads to a correct classification on the basis of Fisher’s linear discriminant functions of 62.8%. More than 60% of
the analyzed presale reports are classified correctly. As outlined in Table 4-6, the detected dominant drivers for classification function are to be seen in “cash flow structure” and the size structure of the tranches. Adding different characteristics in relation
to size structure (e.g. Class A) is of particular value. As outlined by Fender and Kiff
(2005) the assigned rating of a specific tranche interacts both with the tranche’s seniority and the applied rating methodology. Accordingly, senior tranches should be analyzed (and rated) on the basis of an EL approach. Since the expected loss approach is
only applied by Moody’s, we find an economic explanation why it is in the issuer’s
benefit to rely on Moody’s analyzing transactions with high portions of senior tranches
(e.g. since this will lead to higher ratings for large portions of the overall transTable 4-5: Discriminant Analysis and Classification of Set I
Discriminant
Analysis
I
II
III
Eigenvalues
Function
Eigenvalue
% of
Variance
Canonical
Correlation
Test of
Functions
WilksLambda
Chisquare
Significance
1
0.527
78.5
0.587
1 through 2
0.572
124.772
0.000
2
0.145
21,5
0.355
2
0.874
30.182
0.000
1
0.360
71.9
0.514
1 through 2
0.645
98.418
0.000
2
0.140
28.1
0.351
2
0.877
29.456
0.000
1
0.207
100.0
0.414
1 through 2
0.829
42.813
0.000
1
0.359
73.4
0.514
1 through 2
0.651
96.822
0.000
2
0.130
26.6
0.339
2
0.885
27.608
0.000
1
0.215
76.5
0.420
1 through 2
0.772
58.724
0.000
2
0.066
23.5
0.249
2
0.938
14.504
0.000
1
0.307
81.0
0.484
1 through 2
0.714
76.532
0.000
2
0.072
19.0
0.259
2
0.933
15.802
0.000
2
IV
V
VI
Classification
(in %)
Wilks’ Lambda
62.8
60.6
51.1
2
59.3
57.1
61.9
52
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
actions). This argument is supported by the performed comparison of subgroups (Table 4-1), which shows that transactions analyzed by Moody’s have the highest portion
of super senior tranches.
Each variable that is identified by the KS-test as separating the two subgroups at least
one time is integrated in discriminant analysis II. With a classification outcome of
60.6%, the results are slightly lower than those observed in discriminant analysis I.
However, the preselection of variables on the basis of the univariate tests proves to be
quite reasonable. In addition, we adjust the composition of discriminant analyses III to
V in order to reduce the number of dependent variables. Following this approach, the
Table 4-6: Discriminant Analysis of Set I (sorting by Rating Agencies)
Discriminant
Analysis
I
Independent
Variables
Asset Management
Cash Flow Structure
Currency
Maturity
# of Tranches
Transaction Type
Volume
Super Senior
Class A
Others
;ŽŶƐƚĂŶƚͿ
II
Asset Management
Currency
Maturity
# of involved Parties
# of Tranches
Class A
Others
Equity
;ŽŶƐƚĂŶƚͿ
III
Maturity
;ŽŶƐƚĂŶƚͿ
IV
Asset Management
Currency
Maturity
# of Tranches
Class A
Others
;ŽŶƐƚĂŶƚͿ
V
Currency
Maturity
;ŽŶƐƚĂŶƚͿ
VI
Currency
Maturity
Others
;ŽŶƐƚĂŶƚͿ
* Fisher’s linear discriminant functions
Canonical Discriminant
Function Coefficients
Standardized Canonical
Discriminant
Function Coefficients
Function 1 Function 2
Function 1 Function 2
0.8263
2.6646
0.3274
0.0335
0.0093
0.0479
-0.0002
1.2610
-0.7225
97578
-47780
1.0686
0.2588
0.0516
0.0155
-0.0173
-0.9005
9.8033
-0.2850
-2.2989
0.0690
-1.4370
1.0874
0.2412
0.0519
-0.0188
-0.9027
9.8232
-2.2316
0.4354
0.0664
-2.0562
0.2268
0.0546
1.1564
-1.7053
-0.6777
-0.3270
1.1476
0.0058
0.0943
-0.2673
-0.0001
-0.8991
1.1153
-2.0501
-1.2829
-0.9651
1.1798
0.0018
0.0987
0.1362
2.4019
-1.3214
-3.4621
-3.3732
0.3204
0.5792
0.1743
0.4874
0.0216
0.0381
-0.3658
0.2997
-0.1827
0.4692
-0.2627
-0.0710
0.6109
0.0856
0.2167
-0.2126
-0.1522
-0.2137
0.2821
-0.0985
0.4143
0.1377
0.7511
0.0271
-0.0398
-0.2278
0.4714
-0.0181
-0.3742
0.6280
0.0269
0.1724
0.3130
0.6076
-0.0635
-0.2202
1.0000
-0.8466
1.1576
0.0052
0.1272
2.5951
-0.9934
-3.3748
1.8280
-0.0177
-2.4299
1.8394
-0.0014
-4.7812
-2.6991
0.4216
0.1284
0.7544
-0.0432
-0.2283
0.4723
-0.3282
0.6162
0.0765
0.2923
0.6564
-0.0477
0.2318
0.9664
0.9731
-0.2584
0.1207
0.7941
0.5561
0.9792
-0.0210
-0.2299
Classification Function
Coefficients*
Fitch
Moody’s
S&P
9.5814
2.6752
56319
0.0872
3.3844
96926
-0.0009
2.1344
3.1200
-2.5214
-6.1998
1.0437
7.6197
0.0839
1.9783
1.9402
6.6180
-4.0450
3.1167
-2.9735
0.1453
-3.4999
1.0443
5.6056
0.0657
2.1535
6.0823
-4.6305
-2.2566
5.2694
0.1400
-7.4313
5.2350
0.1379
1.5657
-7.7141
8.4252
2.2182
4.5319
0.0252
3.3250
9.7289
-0.0004
1.9518
3.1974
-4.1565
-5.2756
9.2434
6.7491
0.0062
1.9146
1.9101
6.9712
-5.4504
3.3015
-2.4876
0.0633
-1.7886
9.1723
4.7810
-0.0133
2.1301
6.3748
-6.0474
-1.7834
4.5612
0.0587
-4.7343
4.5618
0.0587
-0.2602
-4.7344
8.0454
2.2663
5.6819
0.0404
3.4144
9.4975
-0.0006
1.9063
3.2787
-4.0584
-5.4512
86.946
7.8853
0.0232
2.0078
2.0272
8.8600
-5.2776
2.9822
-2.7797
0.0902
-1.6298
8.7495
5.8533
0.0060
2.2350
8.3657
-5.8526
-2.0614
5.8069
0.0843
-6.4041
5.8017
0.0840
2.3421
-6.4104
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
53
function of discriminant analysis III only contains variables that proved to have separation power in each of the three KS-test sessions. The range of possible variables is
reduced in this context to one single coefficient: “maturity”. However, even with only
one dependent variable, the results of discriminant analysis III are also promising, with
a classification level of 51.1%. A presorting relying on the results both of the KS-test
and the ANOVA delivers the set for discriminant analysis IV. We take the separating
variables found by the application of the ANOVA. We then only include those variables as factors for discriminant analysis IV that also indicate separation power at
least in one section of the KS-test. On the basis of this presorting, we end up with six
different variables, which in turn provide us with a classification of 59.3%. Moving
along this pattern of presorting, we perform discriminant analysis V, including only
those variables of the KS-test which have separation power in the sections for Fitch/
Moody’s and Moody’s/S&P. The derived discriminant analysis consists of the variables “currency” and “maturity” and bears a classification level of 57.1%. If we add
to this constellation the variable with the highest explanatory power in discriminant
analysis IV, we receive the results to discriminant analysis VI or a slightly higher classification of level 61.9%.
The discriminant analysis for Set I (rating agencies) provides us with a first assessment
of the multivariate separation power of the transactions’ characteristics. We find empirical support for the existence of individual patterns of characteristics within each
subgroup (rating agency). Therefore, we can reject the null hypothesis.
Even if the individual classification level of each discriminant analysis is rather high,
we want to further improve the classification level between the different subgroups in
the following chapter. A detailed analysis of the classification matrix for the different
discriminant analyses provides us with some iterating patterns. As exemplarily outlined in Table 4-7 for discriminant analysis V, we observe most allocation errors for
the predicted subgroup Fitch in relation to S&P during the classification process. Presale reports originally prepared by S&P are falsely allocated to the Fitch subgroup. In
the case of discriminant analysis V this is also reciprocally true for the predicted S&P
subgroup classification.
54
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
Table 4-7: Classification of Results of Discriminant Analysis V
Rating Agency
Original
Predicted Group Membership
Total
Fitch
Moody’s
S&P
Original
Count
Count
Fitch
Moody’s
S&P
30
3
20
13
40
25
22
16
62
65
59
107
%
Fitch
Moody’s
S&P
46.2
5.1
18.7
20.0
67.8
23.4
33.8
27.1
57.9
100.0
100.0
100.0
Trying to capture this error term, we perform a multivariate analysis by using the variable “rating methodology” in the following chapter. The subgroup of the “PD approach” consists of Fitch and S&P, while the “EL approach” only incorporates the ratings of Moody’s. This allows us to neutralize the above-mentioned error term and increase the level of classification for our targeted separation of the presale reports.
4.5.2.2 Set II (sorting by Rating Methodologies)
The variable “rating methodology” is the grouping variable for the discriminant analyses of Set II. Similar to the sorting by “rating agencies”, we perform several discriminant analyses relying on the results of the univariate tests with the grouping variable
“rating methodology”. In contrast to the sorting by rating agencies, the sorting by rating methodologies differentiates only between two subgroups. This reduction makes
the application of a discriminant analysis less complex. Table 4-8 shows the results of
the discriminant analyses for different coefficient combinations of the grouping variable “rating methodology”. The detailed outcomes are summarized in Table 4-9. The
significance on the basis of Wilks-Lambda is equal to zero for all realized discriminant
analyses. Therefore, they are all highly significant with regards to their separation
qualities.
The coefficients of discriminant analysis I are derived from the ANOVA’s results of
Set I. Each variable which proves to have univariate separation power under ANOVA
was selected. The combination of the eight variables selected in this way seems to
classify properly, according to the results of Fisher’s linear discriminant function
(77.1%). Exclusion of the standardized canonical discriminant function coefficients
with the least explanatory power (discriminant analysis II) leads to a comparably high
level with 77.1%.
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
55
Table 4-8: Discriminant Analysis and Classification of Set II
Discriminant
Analysis
I
II
III
IV
V
VI
VII
VIII
IX
X
XI
Eigenvalues
Classification
(in %)
Wilks’ Lambda
Function
Eigenvalue
Canonical
Correlation
Test of
Function
WilksLambda
Chisquare
Significance
1
1
1
1
1
1
1
1
1
1
1
0.225
0.225
0.162
0.147
0.161
0.161
0.174
0.181
0.189
0.198
0.198
0.429
0.429
0.373
0.358
0.373
0.373
0.385
0.391
0.399
0.406
0.406
1
1
1
1
1
1
1
1
1
1
1
0.816
0.816
0.861
0.872
0.861
0.861
0.852
0.847
0.841
0.835
0.835
45.736
45.806
34.001
31.324
33.986
33.914
36.331
37.596
39.133
40.713
40.624
0
0
0
0
0
0
0
0
0
0
0
77.1
77.1
77.1
74.9
78.8
78.8
78.8
79.2
79.2
79.7
79.7
The selection process of function coefficients for discriminant analysis III solely relies
on the KS-test of Set II. All variables, which proved to separate the groups “EL approach” and the “PD approach”, are included in the discriminant analysis III. The result in terms of the classification level (77.1%) is exactly the same as already observed
for discriminant analyses I and II.
In order to assemble the function coefficients for discriminant analysis IV, we use the
results of both the ANOVA and the KS-test. We include these variables as function
coefficients that have univariate separation power both in the ANOVA and in the KStest. Following this selection process, we are able to identify the two variables of “currency” and “maturity”. With 74.9%, the relevant discriminant analysis has a slightly
lower level of classification than observed in the preceding analyses. Since these two
variables have the highest univariate separation power, we take them as a starting point
to optimize the classification level for the grouping variable “rating methodology”. In
the following paragraph, we iteratively add one function coefficient to the existing
combination of function coefficients and compute an individual discriminant analysis
after each addition. If the inclusion of a new variable provides a higher or equally high
level of classification, it is added to the set of coefficients.
We start this optimization process by adding a third functional coefficient to the variables of discriminant analysis IV. For each of the remaining fourteen variables, an
individual discriminant analysis is now computed. The highest level of classification
(78.8%) is achieved by adding the variable “Class D” to the variables “currency” and
“maturity”. The results of this analysis are documented in Table 4-8 as well as Table
56
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
4-9 under discriminant analysis V. Trying to achieve an even higher level of classification, we add a fourth variable to the three variables “currency”, “maturity” and “Class
D”. We again compute an individual discriminant analysis for each of the thirteen remaining variables. The accrued results all lead to lower levels of classification. Only
the discriminant analysis including the variable “Class E” achieved an equally high
level of classification with 78.8% (discriminant analysis VI). The inclusion of “transaction type” as a fifth variable during the next optimization step delivers a classification level of 78.8% for discriminant analysis VII that is as high as the optimal discriminant analysis of four variables.
In the next step we add a sixth variable. Again, the individual inclusion of the variables “Class C” and “equity” increases the level of classification up to 79.2% (discriminant analysis VIII and IX). Concerning the inclusion of a seventh variable we
compute the outcomes of the corresponding classification levels. With a classification
level of 79.7% the additional inclusion of the variable “other tranches” leads to the
highest classification level for seven function coefficients (discriminant analysis X).
By adding the eighth coefficient “number of tranches”, we achieve a comparably high
level of classification in discriminant analysis XI. In addition, the inclusion of a ninth
coefficient does not provide a higher level of classification in any case. Following our
optimization approach, the classification levels of discriminant analysis X and XI are
the highest. Since the inclusion of a ninth coefficient does not lead to a higher classification level, we define our optimum in discriminant analysis X. This optimum is the
result of the computation of over 90 discriminant analyses based on the results of the
univariate tests.
The discriminant analysis of Set II reveals high levels of classification. Therefore, we
can state that presale reports of the same rating methodology show an identical pattern
of characteristics, and that we can reject the null hypothesis. The results of the discriminant analyses of Set II show that the grouping variable “rating methodology” leads
to higher classification levels than those obtained under the grouping variable “rating
agency” in Set I. Based on a coefficient selection induced by the univariate test session, the results of discriminant analyses I to IV are only slightly lower than our optimum in discriminant analysis X.
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
57
Table 4-9: Discriminant Analysis of Set II (sorting by Rating Methodologies)
Discriminant
Analysis
Independent
Variables
Canonical Discriminant
Function Coefficients
Standardized Canonical
Discriminant
Function Coefficients
Classification
Function Coefficients
EL Approach PD Approach
I
II
III
IV
V
VI
VII
VIII
IX
Cash Flow Structure
Currency
Maturity
Number of Tranches
Transaction Type
Volume
Super Senior
Others
;ŽŶƐƚĂŶƚͿ
Cash Flow Structure
Currency
Maturity
Number of Tranches
Transaction Type
Volume
Others
;ŽŶƐƚĂŶƚͿ
Currency
Maturity
Class A
Equity
;ŽŶƐƚĂŶƚͿ
Currency
Maturity
;ŽŶƐƚĂŶƚͿ
Currency
Maturity
Class D
;ŽŶƐƚĂŶƚͿ
Currency
Maturity
Class D
Class E
;ŽŶƐƚĂŶƚͿ
Currency
Maturity
Class D
Class E
Transaction Type
;ŽŶƐƚĂŶƚͿ
Currency
Maturity
Transaction Type
Class C
Class D
Class E
;ŽŶƐƚĂŶƚͿ
Currency
Maturity
Transaction Type
Class D
Class E
Equity
;ŽŶƐƚĂŶƚͿ
1.4850
1.0733
0.0258
0.0696
-0.1888
-0.0002
-0.1749
4.3141
-3.6762
1.5076
1.0761
0.0252
0.0665
-0.2095
-0.0002
4.4890
-3.6290
1.1208
0.0456
0.8443
-2.6976
-3.0279
1.2335
0.0495
-2.9251
1.2161
0.0471
8.5303
-3.1736
1.2129
0.0470
8.7283
-0.3645
-3.1655
1.1831
0.0394
8.5230
0.4133
-0.3628
-2.1728
1.1863
0.0397
-0.3604
5.0639
5.6998
-0.3376
-2.3199
1.1330
0.0351
-0.3657
9.9097
1.1493
-4.6482
-1.7513
0.3401
0.5736
0.3959
0.1624
-0.1498
-0.3270
-0.0418
0.2149
0.3453
0.5751
0.3871
0.1551
-0.1663
-0.3510
0.2236
0.5990
0.6993
0.2163
-0.1712
0.6592
0.7581
0.6499
0.7217
0.2939
0.6482
0.7203
0.3007
-0.0113
0.6323
0.6039
0.2936
0.0128
-0.2880
0.6340
0.6084
-0.2861
0.2260
0.1963
-0.0104
0.6055
0.5377
-0.2903
0.3414
0.0356
-0.2950
24.1763
5.9523
0.0158
2.6382
7.1640
0.0009
-10.8782
-74.4865
-33.7175
25.5479
6.1004
-0.0205
2.4410
5.8811
-0.0002
-63.7180
-32.2254
4.1934
0.0744
9.3099
3.10467
-8.6010
4.5849
0.0586
-4.7490
4.7211
0.0579
31.5857
-5.3384
5.0501
0.0668
10.5522
38.6646
-5.7637
4.8516
0.1400
9.0078
29.0813
4.5502
-11.5698
4.9614
0.1445
4.5304
21.1449
-2.0854
25.9789
-11.9072
4.8616
0.1534
4.6372
0.4838
25.3793
22.8583
-12.5336
25.7860
7.1158
0.0438
2.7138
6.9594
0.0006
-11.0678
-69.8100
-36.3451
27.1815
7.2664
0.0069
2.5131
5.6541
-0.0004
-58.8539
-34.8005
5.2220
0.1163
10.0848
28.5709
-10.1040
5.6658
0.1020
-6.0541
5.8357
0.1011
39.4035
-69.715
6.1618
0.1010
18.5517
38.3306
-7.3895
5.9784
0.1776
17.1246
29.4749
4.2047
-12.3472
6.1137
0.1831
4.1803
26.0635
3.4510
25.6509
-12.8600
5.9864
0.1883
4.2741
10.3213
26.5203
18.2439
-12.9612
58
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
Table 4-9 – Continued
X
Currency
Maturity
Transaction Type
Class C
Class D
Class E
Others
;ŽŶƐƚĂŶƚͿ
XI
Currency
Number of Tranches
Maturity
Transaction Type
Class C
Class D
Class E
Others
*Fisher’s linear discriminant functions
1.0953
0.0344
-0.3479
4.8601
6.1630
-4.4796
6.3309
-2.1213
1.0941
0.0032
0.0344
-0.3465
4.8421
6.1701
-4.4919
6.2405
0.5853
0.5278
-0.2762
0.2169
0.2123
-0.1390
0.3154
0.5847
0.0076
0.5266
-0.2751
0.2161
0.2125
-0.1394
0.3109
5.0360
0.1512
4.5365
21.1125
-3.4364
33.8690
-12.0174
-12.0609
4.3350
20.1482
0.1017
5.4081
10.0953
1.0217
26.2330
-67.5935
6.1484
0.1862
4.1831
26.0485
2.8228
29.3195
-5.5877
-12.8932
5.4462
2.0181
0.1366
5.0561
15.0130
7.2882
21.6709
-61.2556
4.5.3 Interpretation
The variable “maturity” and “currency” have a great impact in separating the two
groups of Set II. The discriminant analysis of Set I also confirms the impact of these
two variables. Since we are not able to obtain the weighted average maturity of the
CDO’s asset pool for all analyzed transactions, we do not integrate this characteristic
in our empirical analysis. However, assuming that the correlation between the detected
weighted average maturities and the corresponding legal maturities is reasonably high,
we can use the legal maturity as a proxy for the weighted average maturities. In addition, the weighted average maturity of the asset pool may influence the rating outcome
due to different concepts of modeling default correlations. Only S&P calculates the
recovery rates on the basis of so-called tired recovery rates, assuming not only default
correlation but also a correlation in terms of recovery rates (e.g. a high number of defaults also triggers lower recovery rates than experienced otherwise). Therefore, the
application of tired recovery rates corresponds to lower recovery rates resulting in
higher expected losses since long weighted average maturities correspond to long maturities on the individual credit level. Together with the assumption that long weighted
average maturities correspond to comparably long legal maturities, this supports our
empirical results that the characteristic “maturity” acts as a variable with high discriminatory power in chapter 4.5.2.1. Specifically transactions with long maturities
should – from an investor’s point of view - not be rated by S&P. Following this argu-
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
59
ment, we can identify a rationale why an issuer has an interest not to publish all solicited ratings and rather select the favorable ones.
Additionally, “currency” contains high degrees of explanatory power in both sets. We
can explain these findings in economic terms, since “currency” is generally linked to
the geographical roots of the transaction’s asset pool (e.g. the asset pool of notes denominated in EUR originates from European companies). The asset pool’s geographical affiliation in turn determines its recovery rates. Since recovery rates not only vary
between different regions but also between different rating agencies, we find an economic explanation for the high explanatory power of the variable “currency”. By comparing the medians of the characteristics for each rating agency (see Table 4-1), we
find additional support for this explanation, since the presale reports published by
Moody’s are primarily denominated in EUR, whereas presale reports of S&P mostly
descend from the USD area.
A third characteristic which significantly impacts the classification process is the
transaction’s tranche structure. To optimize the separation power of our discriminant
analysis, we successively include the different tranches (in % of the overall transaction
volume) as discriminant factors. Classifying on the basis of different rating agencies
leads to high discriminatory power for the senior tranches of CDO transactions (see
classification level of Set I). In turn, the inclusion of the more senior tranche portions
(Super Senior, Class A and Class B) does not improve the classification results of Set
II and are therefore excluded. However, the inclusion of mezzanine and junior debt
(Class D notes and below) subsequently improves the classification level of Set II,
since these notes add substantial separation power. Apparently, subordination and attachment points of levels of mezzanine and senior debt affect each rating methodology
differently and contribute to diverse rating outcomes. These findings are in line with
Fender and Kiff (2005), who argue that the higher the differences in tranche structure,
the better the assignment of transactions to a specific rating methodology or rating
agency. Since high subordination levels correspond to a high portion of junior tranches
in the transaction structure the findings of Fender and Kiff (2005) deliver a second
economic explanation for the high explanatory power of the tranche structure: CDO
60
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
transactions with high portions of very junior tranches (e.g. the equity portion) should
be rated by Moody’s.
Cantor and Packer (1997) find evidence that on average, third ratings in bond markets
assign higher ratings than the first two rating outcomes and that the policy of rating on
request induces a sample selection bias. We find first signs of empirical evidence for a
sample selection bias in the CDO market as defined by Cantor and Packer (1997) for
corporate credit ratings. In line with our definition of rating model arbitrage and our
empirical results, the findings of Cantor and Packer therefore support our view of rating model arbitrage (if applied to CDO markets): It is reasonable for an issuer to obtain two or more rating(s), but limit publication to the most favorable ones.
4.6 Conclusion
The main objective of this paper is to provide empirical evidence for rating model arbitrage in CDO markets. On the basis of information asymmetry it is argued that issuers of CDO transactions have economic incentives to take advantage of the uneven information distribution between issuers and investors and to perform rating model arbitrage.
Sorted by rating agency and rating methodology, our empirical analysis provides evidence of homogeneity within the groups and heterogeneity between the groups, i.e. we
are able to detect common patterns within the groups. The existence of these patterns
shows that specific transactions are rated by specific agencies and/or methodologies.
In detail, the variables “currency” and “maturity” incorporate the highest explanatory
power in the discriminant analysis. Additionally, sorting the transactions by different
rating agencies reveals consisting patterns between Fitch and S&P. The results allow
us to classify the presale reports according to the used rating methodologies (EL and
PD), because Fitch and S&P both rely on a PD approach. Eventually, we find strong
classification power for variables referring to the seniority structure of the transactions.
Since we looked at all assessable economic factors on CDO transaction level and follow the findings of Fender and Kiff (2005), we do interpret rating model arbitrage as
the only possible explanation for the above detected patterns in transactions’ characte-
Rating Model Arbitrage in CDO Markets: An Empirical Analysis
61
ristics. Thus, our empirical findings support the assumption that rating model arbitrage
exists in CDO markets.
Our results are closely linked to the findings by Fender and Kiff (2005), who describe
the related rating methodologies in their paper and theoretically argue that there are
economic incentives for issuers to perform rating model arbitrage not only across rating agencies, but across rating methodologies. Thick senior tranches are likely to benefit from the EL rating approach by Moody’s, whereas less senior tranches profit from a
PD approach as incorporated in Fitch’s and S&P’s methodology. However, they did
not find empirical evidence for rating model arbitrage in the CDO context. Our empirical results in turn show that transactions with large super senior tranches are more
likely to receive ratings and according presale reports prepared by Moody’s.
To reduce information asymmetry and lower agency costs, investors need to combine
their market power. With combined market power, investors could claim the publication of all assigned ratings or – following our empirical analysis – at least foster the
publication of two presale reports issued by two rating agencies using different rating
methodologies. In such an optimized market setting, an additional significant principal-agent relationship might develop, namely between the investor (principal) and the
rating agencies (agent). If the rating agencies become agents of the investors and the
investor’s influence on the rating process becomes more direct, higher risk premiums
might be enforced.
Further research might focus on the impact of rating model arbitrage on the pricing of
CDO notes in secondary markets, especially as variables referring to the structure of
seniority have strong classification power. CDO notes usually trade at significantly
higher spreads as comparable corporate bonds of the same rating category. Therefore,
rating model arbitrage could be a driver for higher spread levels in CDO markets. Furthermore, the rating and pricing of individual tranches of CDO structures in the context of rating model arbitrage might be an interesting research topic. Additionally, potential arbitrage strategies for investors based on the different rating methodologies
could be analyzed.
62
Impact of Multiple CDO Ratings on Credit Spreads
5 Impact of Multiple CDO Ratings on Credit Spreads
5.1 Introduction
Rating agencies play a prominent role in CDO markets. 2 In order to overcome existing
information asymmetries CDO investors rely on rating agencies as information agents
and use ratings as guidance throughout the investment process. Based on these considerations we analyze the impact of multiple ratings on credit spreads of the respective
CDO tranches. Prior to its issuance CDO tranches are typically rated by one, two or
three rating agencies. According to the assigned rating(s), the issuance tranche spread
is determined. We argue that each additional rating on top of the first one incorporates
new incremental information and thus reduces information asymmetry between the issuer and the investor. Reduced information asymmetry increases transparency, thereby
lowers investors’ demand for risk premiums and leads to lower credit spreads. The
motivation for this empirical analysis becomes especially relevant when considering
the current financial crisis. Among others, information asymmetries between issuers
and investors and misaligned incentive structures for issuers along the structuring
process of CDOs lead to a situation where only insufficient information was shared
with investors.
Our findings are threefold: First, we find that on average credit spreads indeed decrease with an increasing number of ratings. In a regression of the number of outstanding ratings on credit spreads controlling for various factors (e.g. maturity), we document significant impact levels for multiple ratings. We show that in addition to other
pricing factors (e.g. credit quality) the number of outstanding ratings incorporates explanatory power with respect to the pricing structure of CDO credit spreads. Second,
even with decreasing spread levels in place, we were not able to confirm the hypothesis that marginal tranche spread reduction decreases with the number of published ratings. Third, we found that in the case of joint (pair wise) ratings, on average Fitch assigned a higher credit quality (e.g. better rating) than its competitors Moody’s and
S&P for the very same CDO tranche. Since Fitch is by far the smallest of the three rat-
2
The following chapter represents joint work by Morkötter and Westerfeld (2009a).
Impact of Multiple CDO Ratings on Credit Spreads
63
ing agencies in the field of structured credit, we see a potential explanation in the form
of a selection bias.
Several papers have been published on corporate ratings to analyze the impact of multiple ratings on credit spreads and implied risk premiums. However, the application on
or transfer to CDOs cannot be found in finance literature yet. Accordingly, we analyze
if CDO markets value credit ratings and, in particular, if and how the mere existence
of multiple ratings from different rating agencies is priced into credit spreads of CDO
tranches. Earlier studies argue that issuers of CDOs have economic incentives to take
advantage of uneven information distribution between issuers and investors and label
it rating model arbitrage (Fender and Kiff, 2005). At the present time, investors do not
appear to be teaming up to enforce publication of multiple CDO ratings but rather accept rating model arbitrage in CDO markets. However, investors might add a risk
premium for information asymmetry and potential rating model arbitrage when pricing
CDO tranches. Against this background, we argue that additional ratings in turn should
lower this risk premium leading to ceteris paribus lower tranche spreads. In addition,
by introducing a double-step interpolation process to create a benchmark index for
CDO spreads we also extend existing literature on pricing factors of CDO tranches in
terms of applied methodologies.
The paper is organized as follows: Based on a literature review in chapter 5.2, we
present the specific characteristics of the CDO rating market and accordingly formulate three different hypotheses (chapter 5.3). The underlying data sample for the empirical part is introduced in chapter 5.4, as well as the description of index-adjusted credit
spreads. Along the defined hypotheses we eventually analyze the data in chapter 5.5
and perform several tests including a multiple regression analysis. Chapter 5.6 concludes the paper.
5.2 Literature Review
Two very recent papers analyzing credit spreads of CDOs are Schiefer (2008) and
Vink and Thibeault (2008). The latter compares credit spreads within different segments of the securitization market. The authors find that pricing factors differ significantly between CDOs, asset-backed securities and mortgage-backed securities. Be-
64
Impact of Multiple CDO Ratings on Credit Spreads
sides their detailed analysis they do not address the topic of multiple ratings. Schiefer
(2008) additionally provides a comprehensive analysis of pricing factors for CDO credit spreads. Again, the paper’s focus is not on the role of multiple ratings within CDO
markets but on pricing factors in general. However, in the course of our empirical
analysis we were able to confirm some of their basic results for a larger data set of
CDO tranches.
With regard to corporate bonds, various studies have been published dealing with multiple ratings. Generally, two types of literature can be distinguished: (1) analyzing the
question why borrowers obtain more than one rating; and (2) assessing the impact of
multiple ratings on bond yields.
Referring to (1), Cantor and Packer (1995) analyze whether the reason for getting an
additional rating may be regulatory in nature. Many financial institutions have limits,
either self imposed or imposed by government regulators, on the amounts of debt they
can hold of certain ratings. As most of these regulations only require that the highest or
second highest rating be above the cutoff point, the firm’s chances of meeting the
standard increase if a third or fourth rating is obtained. Therefore, issuers could have a
strong incentive to obtain multiple ratings to reach those investors. However, the authors find no evidence that firms obtaining Fitch IBCA ratings are doing so in order to
meet rating regulation requirements. In a later paper, Cantor and Packer (1997) empirically test for the existence of rating model arbitrage in bonds. They find evidence that
third ratings in bond markets on average assign higher ratings than the first two rating
outcomes and that the policy of rating on request induces a sample selection bias.
However, they find no evidence for the theories that only firms with greater default
risk uncertainty or firms engaged in rating shopping are interested in obtaining third
ratings.
In contrast to bonds, CDO rating methodologies applied by the major three rating
agencies differ substantially, which can result in clear differences in the ratings assigned by the agencies to certain tranche structures (Peretyatkin and Perraudin, 2002).
Moody’s has long relied on an EL criterion, as opposed to a criterion that focuses primarily on PD, as applied by its competitors S&P and Fitch. Other things being equal,
an EL approach may therefore be more favorable to large senior tranches than a de-
Impact of Multiple CDO Ratings on Credit Spreads
65
fault probability approach, and less favorable towards more junior tranches that tend to
be of thinner size. Fender and Kiff (2005) explore the impact of differences in methodologies across rating agencies for senior tranche rating outcomes. They conclude that
because investors do not fully understand the possible implications of the effects analyzed for tranche ratings, rating model arbitrage is a theoretical possibility. In practice,
the authors could only find limited evidence for this behavior. However, Morkötter
and Westerfeld (2009b) find no empirical evidence to accept the hypothesis stating
non-existence of rating model arbitrage on the basis of information asymmetry, as patterns of transaction characteristics per rating agency/rating methodology could be
identified.
The second branch of literature deals with the effect of ratings on bond yields. These
papers add to the question why borrowers seek a third or fourth rating as these ratings
might convey information to the markets that reduces the cost of borrowing for the issuers. For the purpose of this paper, we explicitly differentiate between split and multiple ratings. In finance literature split ratings are defined as (bond) ratings in which
two or more rating agencies assess the very same financial product but come up with
different ratings (e.g. Jewell and Livingston, 1998). Multiple ratings in turn refer to the
mere number of ratings existing for a specific entity/note regardless if the rating results
differ. However, following Jewell and Livingston (1999), the papers fail to reach a
consensus on how the market prices bonds with split ratings. Kish et al. (1999) conclude that the market finds value in the ratings from each agency (Moody’s and S&P)
and that there is not enough evidence that the market values one agency over the other.
Billingsley et al. (1985), Liu and Moore (1987) and Perry, Liu, and Evans (1988) find
that the market prices bonds with split ratings as if only the lower of the two ratings
conveys information. In contrast, Hsueh and Kidwell (1988) and Reiter and Ziebart
(1991) conclude that markets price bonds as if only the higher of the two ratings conveys information. Jewell and Livingston (1998) conclude that when firms receive a
split rating from Moody’s and S&P, the markets considers an average of the two ratings when determining default spreads for the bond. Thus, markets place some value
on both bond ratings. In a later paper, Jewell and Livingston (1999) even show that the
bond markets value the ratings of three raters. The authors compare bond ratings of
66
Impact of Multiple CDO Ratings on Credit Spreads
Fitch to those of Moody’s and S&P to analyze the potential benefits of seeking out additional ratings from a smaller rating agency (Fitch), by comparing rating levels, rating
changes, and the impact of ratings on bond yields. Inter alia, the authors test for the
hypothesis that the average observed rating from Fitch is likely to be significantly
higher than the “true” average rating from the two other agencies. Their analysis confirmed this hypothesis.
Like Jewell and Livingston (1999) in their analysis for bonds, we are not concerned
with the determinants of ratings like the first branch of literature, as we accept the fact
that CDO markets are multi-rating markets. However, we take the possibility for rating
shopping based on information asymmetries between originators and investors as a
given and accept this as a potential reason for adjusted risk premiums. Also, unlike the
second branch of literature, we are not concerned with future default. Rather, we take
the ratings as a given and analyze the market perception of the number of available ratings in terms of credit spreads.
5.3 Multiple Ratings and Credit Spreads within CDO Markets
As a starting point, we develop a basis outline of the CDO rating market: A plain vanilla CDO transaction is typically centered on a SPV, which invests into various credit-linked assets (e.g. SME loans, bonds or tranches of other CDO transactions) and
refinances its purchases through the issuance of notes, i.e. CDO tranches. Subordinated tranches act as credit enhancement for the more senior notes (subordination
principle). The tranches are bought by international investors, e.g. hedge funds, banks
or other SPVs investing in CDOs with each note paying interest either defined as a
fixed rate or as a spread premium over a certain reference benchmark (typically some
sort of LIBOR). According to market standards, the issuer (investment bank or external asset manager) assigns one or more rating agency to assess the transaction and
provide a rating for all or some of the underlying tranches prior to note issuance. The
issuer will only be willing to publish and share this information with investors if (i) the
ratings are favorable for the tranches or if (ii) investors explicitly request the rating.
The CDO rating market is an oligopolistic one with Fitch, Moody’s and S&P as the
three dominating players.
Impact of Multiple CDO Ratings on Credit Spreads
67
Ratings incorporate signaling attributes and are therefore used as a marketing instrument. Since CDO ratings are typically solicited, the issuer has to cover the costs. On
average, these costs are around 4.25 bps (Standard & Poor’s, 2007) of the issuance volume. Consequently, the more rating agencies are assigned, the higher the underlying
cost of the note issuance.
CDO rating processes diverge significantly from the rating process of corporate bonds.
Unique features of the CDO rating market are the accessibility of rating tools used by
the agencies, the different methodologies used (EL approach by Moody’s, PD approach by S&P and Fitch) and the close cooperation between agency and issuer during
the negotiation phase. The latter are heavily discussed in the wake of the recent subprime crisis. We do not intend to discuss independency issues of rating agencies here;
however, we proceed with the assumption that the relationship and exchange between
rating agencies and issuers is very close and thereby impacts information efficiency. In
comparison to the bond rating market, a specific feature of the CDO rating market is
the fact that S&P and Fitch apply the same rating methodology (PD-based) whereas
Moody’s relies throughout the rating process on an EL-based approach.
As a basis for our empirical analysis, we now formulate a set of different hypotheses.
The first hypothesis is related to the fundamentals of information asymmetry and focuses on the principal-agent relationship between investor (principal) and issuer
(agent): We argue that the credit spread of an individual CDO tranche is impacted by
the number of outstanding ratings. The publication of a rating impacts the information
distribution between the issuer and the investor. Thus we assume that each additional
tranche rating conveys incremental information. Reduced information asymmetry
should therefore lead to ceteris paribus lower credit spreads since the investors demand
a lower premium due to less uncertainty about the credit quality. We test our first hypothesis as follows:
68
Impact of Multiple CDO Ratings on Credit Spreads
Hypothesis H10: The number of outstanding ratings does not influence the credit
spreads of CDO tranches.
Hypothesis H11: The number of outstanding ratings does influence the credit
spreads of CDO tranches (negatively).
In addition, we further investigate the magnitude of tranche spread reduction in absolute numbers. In particular, we compare the tranche spread reduction from single to
double ratings and from double to triple ratings. Diminishing marginal utility should
reduce the value of incremental information provided by each additional rating. Therefore, one could expect tranche spread reduction from single to double ratings to be
larger than from double to triple ratings. However, the reduction of tranche spreads
can also be the mere result of a selection bias. Since the issuer decides which rating is
published and investors demand two ratings on average, the issuer will only publish a
third rating if the rating outcome is favorable to the transaction. Thus one could expect
spread reduction from double to triple ratings to be larger than from single to double
ratings. Based on these considerations, we hypothesize that:
Hypothesis H20: Additional (marginal) ratings lead to a decreasing or constant reduction of the tranche spread.
Hypothesis H21: Additional (marginal) ratings lead to an increasing reduction of
the tranche spread.
Based on Jewell and Livingston (1999) we additionally analyze the role of Fitch within CDO markets. In particular, we test the hypothesis that the average rating by Fitch
is considerably higher than the average rating by Moody’s and S&P. Fitch as the smallest of the three rating agencies might try to capture market share through the issuance
of ratings in favor of the transaction. A second explanation again centers on the existence of a potential selection bias. We argue that Fitch is only considered by the issuer
in case the rating outcome is expected to be favorable to the transaction. Thus, we propose:
Impact of Multiple CDO Ratings on Credit Spreads
69
Hypothesis H30: The average Fitch rating is not different from the average S&P or
Moody’s rating.
Hypothesis H31: The average Fitch rating is different (and better) than the average
S&P or Moody’s rating.
In order to eliminate potential bias in the empirical results we perform several robustness checks, including control variables for split ratings as well as for tranches solely
rated by Fitch, Moody’s or S&P.
5.4 Data Sample
We started our data analysis by setting up a database provided by Deutsche Bank consisting of a unique set of 9,536 CDO tranches from 1,454 transactions issued between
January 2004 and March 2007. From these 9,536 CDO tranches, we removed all
tranches with missing data points (e.g. rating or credit spread). For reasons of consistency we also excluded all tranches with fixed credit spreads as this would have caused
dilution of our empirical results when approximating spread levels. Thus, our data
sample incorporates only CDO tranches with floating credit spreads (e.g. three-month
LIBOR plus X).
Accordingly, we used a total data set of 5,133 tranches. For each tranche the outstanding rating(s), transaction type, lead manager, issue date, rated volume, as well as the
level of seniority within the specific transaction is available. We distinguish between
three different transaction types (CLO, CBO & Exotic) and use this information as an
independent variable in our regression analysis. As outlined in Table 5-1, the total
number of tranches can be differentiated into tranches that carry single, double or even
triple ratings. We grouped the sample according to rating agencies.
With 4,874 respectively 4,596 rated CDO tranches, S&P and Moody’s are the dominating rating agencies in our sample. Fitch has the smallest share with 1,281 rated
tranches (24.96%). 271 tranches (5.28%) carry a single rating, 4,106 tranches
(79.99%) have double and 756 tranches (14.73%) carry triple ratings. If we further
analyze double-rated tranches, we see that ratings are predominately published by the
70
Impact of Multiple CDO Ratings on Credit Spreads
Table 5-1: Data Sample “Multiple CDO Tranche Ratings”
CDO Tranche Ratings
Total Sample
Rating Agency
Fitch
Moody’s
S&P
Single Rating
Total
Fitch
Moody’s
S&P
Double Rating
Total
Moody’s/ S&P
Fitch/ S&P
Moody’s/ Fitch
Triple Rating
Fitch/ Moody’s/ S&P
# of rated
Tranches
5,133
in % of
Total Sample
ϭϬϬ͘ϬϬй
Mean
Maturity
(in years)
7.96
Mean
Volume
(in mUSD)
86.38
Mean
Rating
Code
4.98
1,281
4,596
4,874
Ϯϰ͘ϵϲй
ϴϵ͘ϱϰй
ϵϰ͘ϵϱй
7.77
7.97
7.95
100.21
89.92
86.14
4.86
4.92
4.92
271
68
37
166
ϱ͘Ϯϴй
ϭ͘ϯϮй
Ϭ͘ϳϮй
ϯ͘Ϯϯй
7.48
9.20
7.22
6.83
45.20
31.27
47.06
50.49
5.42
7.25
4.65
4.84
4,106
3,649
303
154
ϳϵ͘ϵϵй
ϳϭ͘Ϭϵй
ϱ͘ϵϬй
ϯ͘ϬϬй
8.07
8.08
8.04
7.74
83.81
83.55
64.69
127.53
5.02
5.01
5.14
5.10
756
ϭϰ͘ϳϯй
7.54
115.08
4.55
combination of Moody’s and S&P (71.09%). The combination Fitch/ S&P (5.90%)
respectively Moody’s/ Fitch (3.00%) are ranked second and third. One potential explanation might be the higher market share of S&P in comparison to Moody’s and the
fact that the same rating methodology is applied by Fitch and S&P.
As can already be seen in Table 5-1, mean maturity of the transactions seems to be
higher for double ratings than for single and triple ratings. However, concerning single
ratings, it seems that Fitch rated not only the tranches with the longest maturity, but
also with the lowest volume per tranche. Overall, the volume of single rated tranches is
significantly lower than the volume of double and triple rated tranches. In addition,
single rated tranches are not only the smallest in terms of volume, but also receive the
lowest ratings in comparison to double and triple ratings. Both maturity and volume
are important factors for debt instruments and will be analyzed in detail in chapter 5.5.
Concerning rating outcomes we refer to Table 5-2, which outlines the mapping of the
individual rating notches of Fitch, Moody’s and S&P on a numerical scale based on
underlying one-year default probabilities (derived from Fitch, Moody’s, S&P). This
approach is commonly used in finance literature to be able to compare different rating
scales (see e.g. Cantor and Packer 1995). Thereby, we are able to compare the rating
outcomes of all three rating agencies. In the following, the terms “rating outcome”,
“rating code” as well as “rating notch” are used synonymously.
Impact of Multiple CDO Ratings on Credit Spreads
71
Table 5-2: Mapping Code for the Individual Rating Notches
Code
Fitch
Class
Moody’s
Class
S&P
Class
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
AAA
AA+
AA
AAA+
A
ABBB+
BBB
BBBBB+
BB
BBB+
B
BCCC+
CCC
CCCCC
C
D
Aaa
Aa1
Aa2
Aa3
A1
A2
A3
Baa1
Baa2
Baa3
Ba1
Ba2
Ba3
B1
B2
B3
Caa1
Caa2
Caa3
Ca
AAA
AA+
AA
AAA+
A
ABBB+
BBB
BBBBB+
BB
BBB+
B
BCCC+
CCC
CCCCC
SD
D
As outlined in Table 5-1, our data set with 5,133 different CDO tranches incorporates
a large portion of multiple ratings, which can be grouped according to the rating outcome. In the case of identical ratings per CDO tranche, all involved rating agencies
assign the very same rating code for a specific tranche and calculated notch differences
are zero. Split ratings occur only if the rating codes deviate from each other. Thus, in
case of split ratings, notch differences are always unequal to zero. Based on calculation of notch (code) differences credit quality assessments by the agencies can be
compared. According to the number of involved rating agencies (double or triple ratings), one or three notch differences can be calculated. We refer to these rating pairs as
joint ratings, with split ratings being a specific form of joint ratings, but not the only
one.
Table 5-3 displays the notch differences for our data set. Due to triple ratings the total
number of joint ratings exceeds the number of 5,133 analyzed CDO tranches (448 split
ratings in total). The level of differences – as documented in Table 5-3 – shows a di-
72
Impact of Multiple CDO Ratings on Credit Spreads
Table 5-3: Notch Differences of jointly-rated CDO Tranches
# of joint ratings
Rating Differences
# of identical
# of one notch
# of two notches
# of three notches
# of four notches
Moody’s ./. S&P
4,405
Fitch ./. S&P
1,059
Moody’s ./. Fitch
910
Total
6,374
4,126
248
26
5
0
279
987
59
11
2
0
72
813
79
17
0
1
97
5,926
386
54
7
1
448
-39
-0.0368
57
0.0626
# number of Split Ratings
Level of Differences
Overall Difference*
67
Mean*
0.0152
* In case of (-), subtrahend has rated on average lower.
rect comparison of the rating outcomes of each rating agency. For joint ratings, Moody’s ratings are on average lower than the corresponding S&P ratings. In turn, S&P
ratings are on average lower than the Fitch ratings.
This pattern also holds for joint ratings of Fitch and Moody’s. Again, on average Fitch
ratings are better than the corresponding ratings of the second rating agency (e.g.
Moody’s rated lower). It is noteworthy that split ratings mostly lie within a one-notch
range. Our results confirm earlier research in the fields of corporate bond ratings (e.g.
Cantor et al., 1997).
5.5 Empirical Results
5.5.1 Analysis of Credit Spreads
In order to derive a reliable credit spread to capture the effect of multiple ratings several adjustments have to be made. Since we already excluded all CDO tranches with
fixed credit spreads, our data base only consists of floating credit spreads over LIBOR
‫ ݐ݅ܵܥ‬, denominated in bps of the nominal volume of tranche i at the date of issuance t.
In the following we refer to ‫ ݐ݅ܵܥ‬as the unadjusted credit spread.
Then we separate the part of the spread representing the systematic risk of the specific
CDO tranche. By doing so we are able to analyze idiosyncratic credit spreads of different tranches without any dilution from systematic credit risk. We achieve this goal
by subtracting an average CDO Credit Spread Index ‫ ܿݎݐ݆ݔ݁݀݊ܫ ܵܥ‬from the individual
Impact of Multiple CDO Ratings on Credit Spreads
73
unadjusted credit spread ‫ ݐ݅ܵܥ‬and receive an adjusted credit spread of the individual
tranche i:
‫ ݐ݅ܵܥ‬െ ‫ݐ݅ܵܥܣ = ܿݎݐ݆ݔ݁݀݊ܫ ܵܥ‬
In the context of the CDO Credit Spread Index, j refers to a specific sub-index, t indicates the issuance date, r stands for the rating class of the CDO tranche and c for the
currency in which the tranche is denominated. By introducing ‫ ܿݎݐ݆ݔ݁݀݊ܫ ܵܥ‬we are able
to calculate the corresponding adjusted credit spread ‫ ݐ݅ܵܥܣ‬for each individual tranche,
with a specific tranche i and its issuance date t. In the following we refer to ‫ ݐ݅ܵܥܣ‬as
the adjusted credit spread.
For the calculation of ‫ ܿݎݐ݆ݔ݁݀݊ܫ ܵܥ‬we rely on three different sub-indices provided by
Deutsche Bank. Each sub-index refers to a specific class of CDO transactions: CLO,
CBO as well as exotic CDOs (e.g. ABS CDO, CDO Squared). According to the transaction structure, each tranche is flagged to match one of the three sub-indices. We do
not only differentiate between transaction structures, but also between currency denominations, which enlarge the number of sub-indices by a factor two: tranches denominated in USD are attributed to CDX-based sub-indices (one for each transaction type)
and tranches denominated in EUR or other currencies are attributed to iTraxx-based
sub-indices (one for each transaction type).
The six different sub-indices were originally not available for every rating class as presented in Table 5-2. Both for CDX and iTraxx sub-indices are only available for the
rating codes 1, 6, 9 and 12, resulting in 24 sub-indices in total. Furthermore, the original sub-indices calculated by Deutsche Bank do not always refer to the same maturity.
We overcome these two problems by a double interpolation process (used e.g. by
Blanco et al. (2005) for corporate bonds and credit default swaps in order to address
the issue of missing data). In a first step we equalize the term structure of the 24 existing sub-indices and fix the index maturity at ten years. Interpolation of maturity mismatches relies on the term structure of CDS Spreads (3, 5, 7 or 10 years; again divided
into CDX and iTraxx). In a second step we create the missing sub-indices for the rating codes 2 to 5, 7, 8, 10, 11 as well as 13 -16 by a second interpolation (again divided
into CDX and iTraxx). Since our data sample only consists of rating codes between 1
74
Impact of Multiple CDO Ratings on Credit Spreads
and 16, we do not need to compute the sub-indices for the rating codes 17 to 22. We
do not follow the concept of linear interpolation but rely on the mean default probability distribution of the rating agencies as displayed in Table 5-2 in order to interpolate
the missing data points. In the end, this two-step interpolation process leaves us with
two sets (CDX and iTraxx) of sub-indices with each set consisting of three indices
(CLO, CBO and Exotic) for all rating classes (1-16) for the period from January 2004
to March 2007, resulting in 96 different CDO credit spread indices in total. Within this
period sub-indices are calculated for every day on which a price for the basic CDO
credit spreads indices was fixed by Deutsche Bank. Throughout the calculation of
‫( ݐ݅ܵܥܣ‬see formula above) we finally adjust the corresponding CDO credit spread index for the maturity of the corresponding CDO tranche by a process of linear interpolation based on CDS term structure.
In order to give a comprehensive overview of ‫ ݐ݅ܵܥ‬and ‫ ݐ݅ܵܥܣ‬we display the results on
an aggregated level in Table 5-4 and 5-5. We group the underlying credit spreads by
the number of ratings (Table 5-4) and by rating agencies (Table 5-5). For each table
the results are displayed for each rating code in detail (1-16). It shows that in total the
unadjusted credit spread ‫ ݐ݅ܵܥ‬positively correlates with the underlying rating code, i.e.
a decreasing credit quality leads to an increasing spread level. Only for the rating
codes 11 and 13 does this pattern not hold. However, with 64 respectively 57 tranches
these ratings classes are rather small in comparison to the total data sample (5,133
tranches). Despite a few exceptions the assumption of positive correlation between
‫ ݐ݅ܵܥ‬and ‫ ݐ݅ܵܥܣ‬also holds for single, double and triple ratings and rating agencies.
In both Tables 5-4 and 5-5 the unadjusted credit spread is always positive. Due to subtraction of ‫ ܿݎݐ݆ݔ݁݀݊ܫ ܵܥ‬, ‫ ݐ݅ܵܥܣ‬is significantly lower than ‫ ݐ݅ܵܥ‬and in some cases even
negative. Thus, negative ‫ ݐ݅ܵܥܣ‬are a result of the benchmark composition of
‫ ܿݎݐ݆ݔ݁݀݊ܫ ܵܥ‬.
The six benchmark indices do not exclusively display the tranches included in our data
sample but consider more transactions, which were sorted out, i.e. fixed rated tranches.
This mismatch together with the applied two-step interpolation process explains why
‫ ݐ݅ܵܥ‬െ ‫ ܿݎݐ݆ݔ݁݀݊ܫ ܵܥ‬is in total not equal to zero (14.04 bps).
Impact of Multiple CDO Ratings on Credit Spreads
75
Based on the findings in Table 5-5 we note first signs of differences between rating
agencies. Whereas the unadjusted credit spread is on a comparable level (132.07 bps
vs. 134.92 bps) for Moody’s and S&P, it is lower for tranches rated by Fitch (122.99
bps). In the case of adjusted credit spreads we find a comparable pattern (15.63 bps for
S&P, 14.06 bps for Moody’s but only 1.64 bps for Fitch). Homogeneity between the
group of tranches rated by Moody’s and S&P in distinction to tranches rated by Fitch
also holds for tranche volume. In comparison to the average tranche volume of Moody’s (89.92 mUSD) and S&P (86.14 mUSD), Fitch is higher on average (100.21
mUSD).
To test the significance of differences in mean values between the rating agencies we
perform a series of t-tests. As expected, the difference between mean values of the
group of tranches rated by Moody’s and the group of tranches rated by S&P is not significant. In turn, for both groups the difference in mean values compared to the group
of tranches rated by Fitch is significant at the 0.05 level. The only exception is illustrated for the case of tranche volume: the difference between the groups “Fitch” and
“Moody’s” is not significant at the 0.05 level. These test results also hold for a oneway ANOVA performed for the three groups stating the affiliation to a specific rating
agency. For all variables (here also for tranche volume) the group means are not equal
at a significance level of 0.05.
1
79
89.94
6.67
52.53
45.00
0.49
0.00
150.00
18.34
14.97
1.48
-37.64
113.91
1,276
207.23
7.26
35.54
32.00
0.43
0.00
180.00
2.88
-1.55
5.56
-40.48
131.60
267
265.57
7.22
38.74
35.00
0.44
0.00
130.00
6.77
3.33
2.32
-40.54
87.60
1,622
211.12
7.22
36.90
33.00
0.45
0.00
180.00
4.27
0.94
3.98
-40.54
131.60
Total
271
45.20
7.48
170.35
105.00
1.00
0.00
1250.00
22.36
21.31
4.69
-496.84
689.67
4,106
83.81
8.07
136.83
69.00
1.08
0.00
900.00
16.23
3.75
4.09
-498.36
493.31
756
115.08
7.54
111.23
65.00
1.01
0.00
725.00
-0.84
2.44
82.85
-345.32
411.30
5,133
86.38
7.96
134.83
70.00
1.08
0.00
1,250.00
14.04
3.88
4.97
-498.36
689.67
† scaled by absolute mean values of tranche volumes.
Code
Single Rating
# of tranches
sŽůƵŵĞŝŶŵh^;DĞĂŶͿ
DĂƚƵƌŝƚLJŝŶzĞĂƌƐ;DĞĂŶͿ
hŶĂĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
Mean
Median
Standard Deviation†
Minimum
Maximum
ĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
Mean
Median
Standard Deviation†
Minimum
Maximum
Double Ratings
# of tranches
sŽůƵŵĞŝŶŵh^;DĞĂŶͿ
DĂƚƵƌŝƚLJŝŶzĞĂƌƐ;DĞĂŶͿ
hŶĂĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
Mean
Median
Standard Deviation†
Minimum
Maximum
ĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
Mean
Median
Standard Deviation†
Minimum
Maximum
Triple Ratings
# of tranches
sŽůƵŵĞŝŶŵh^;DĞĂŶͿ
DĂƚƵƌŝƚLJŝŶzĞĂƌƐ;DĞĂŶͿ
Unadjusted Credit Spread (ŝŶďƉƐͿ
Mean
Median
Standard Deviation†
Minimum
Maximum
ĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
Mean
Median
Standard Deviation†
Minimum
Maximum
Total
# of tranches
sŽůƵŵĞŝŶŵh^;DĞĂŶͿ
DĂƚƵƌŝƚLJŝŶzĞĂƌƐ;DĞĂŶͿ
hŶĂĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
Mean
Median
Standard Deviation†
Minimum
Maximum
ĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
Mean
Median
Standard Deviation†
Minimum
Maximum
12.33
5.39
1.91
-15.45
110.27
50.55
42.50
0.45
21.00
145.00
86
44.60
8.24
31.01
35.18
0.61
1.29
55.12
65.00
65.00
0.28
41.00
90.00
8
35.68
7.26
9.11
1.86
2.57
-15.45
110.27
48.02
40.00
0.48
21.00
145.00
73
44.95
8.46
29.38
28.60
0.27
21.62
40.83
64.40
57.00
0.20
55.00
85.00
5
53.84
6.60
2
6.60
1.67
3.15
-56.85
159.87
57.52
52.00
0.38
14.00
200.00
865
41.98
8.06
7.30
6.08
3.02
-50.26
98.24
61.06
56.00
0.39
15.00
160.00
130
51.72
7.74
5.16
0.76
3.56
-56.85
119.97
55.77
52.00
0.36
19.00
170.00
696
40.23
8.20
29.98
40.09
1.25
-36.26
159.87
77.00
80.00
0.45
14.00
200.00
39
40.84
6.60
3
7.12
2.31
4.21
-41.14
104.63
73.64
64.50
0.38
18.00
165.00
84
24.77
6.90
1.09
1.93
27.29
-36.54
79.59
67.85
65.00
0.52
18.00
150.00
13
17.86
6.59
3.00
1.71
7.97
-41.14
74.74
69.67
63.00
0.29
40.00
130.00
63
25.36
6.93
49.39
59.43
0.86
-21.43
104.63
114.38
110.00
0.34
35.00
165.00
8
31.38
7.16
4
-12.27
-22.70
4.40
-144.01
125.96
86.95
70.00
0.53
22.00
225.00
56
28.05
7.68
-36.99
-43.53
1.20
-144.01
85.27
71.29
68.50
0.37
25.00
150.00
24
31.66
8.04
-8.22
-4.76
5.66
-84.02
125.96
86.85
69.00
0.57
22.00
225.00
26
27.88
7.70
69.05
69.41
0.50
34.39
122.44
150.00
170.00
0.32
90.00
200.00
6
14.33
6.17
5
15.20
7.18
2.78
-205.00
240.76
111.26
95.00
0.44
24.00
350.00
768
29.35
8.31
-3.45
0.15
14.46
-156.62
175.20
106.26
90.00
0.44
29.00
260.00
90
31.59
7.77
16.96
7.61
2.31
-205.00
240.76
111.18
95.00
0.44
24.00
350.00
642
28.99
8.42
30.49
24.54
1.90
-129.59
187.68
125.03
122.50
0.43
45.00
250.00
36
30.06
7.68
6
10.52
4.62
5.90
-216.62
292.55
136.07
130.00
0.43
60.00
500.00
192
25.79
8.37
-23.81
-45.79
2.35
-167.37
100.83
115.79
100.00
0.37
70.00
225.00
33
19.35
8.23
15.70
5.64
3.77
-216.62
292.55
137.74
132.50
0.44
60.00
500.00
146
27.10
8.44
39.60
42.84
1.98
-63.31
199.02
168.85
165.00
0.41
65.00
325.00
13
27.42
7.91
7
-4.02
-28.58
31.61
-310.95
325.84
194.44
145.00
0.55
55.00
525.00
62
18.25
8.05
-69.53
-71.24
0.86
-246.79
28.02
145.23
140.00
0.25
105.00
265.00
22
24.37
8.87
30.39
20.64
4.64
-310.95
325.84
220.38
150.00
0.56
55.00
525.00
39
14.88
7.62
95.24
95.24
95.24
95.24
265.00
265.00
265.00
265.00
1
15.00
6.70
8
35.48
20.49
2.48
-345.32
490.00
247.31
250.00
0.40
16.00
750.00
773
22.32
8.15
11.78
32.45
8.60
-345.32
352.13
239.07
262.50
0.37
53.00
550.00
110
23.52
7.34
41.50
20.96
1.99
-287.00
490.00
248.72
242.50
0.40
16.00
750.00
622
22.35
8.26
7.77
0.71
14.40
-300.09
270.30
248.05
250.00
0.34
75.00
475.00
41
18.74
8.68
9
-30.88
-58.26
4.31
-498.36
419.40
252.21
200.00
0.52
45.00
800.00
205
21.22
8.54
-123.60
-124.51
0.73
-343.05
61.72
199.56
200.00
0.30
61.00
335.00
27
24.48
8.10
-13.87
-50.13
9.66
-498.36
419.40
260.94
195.00
0.53
45.00
800.00
171
20.74
8.59
-88.71
-75.55
-1.03
-205.91
69.36
242.14
200.00
0.48
175.00
500.00
7
20.26
9.00
10
213.81
259.27
0.82
-386.88
493.31
588.94
625.00
0.25
100.00
900.00
64
9.74
8.27
141.09
243.31
1.49
-197.20
372.94
515.70
587.50
0.28
275.00
650.00
10
6.02
6.90
233.12
271.80
0.71
-386.88
493.31
607.88
625.00
0.23
100.00
900.00
52
10.03
8.59
75.25
75.25
2.42
-53.31
203.81
462.50
462.50
0.42
325.00
600.00
2
20.90
6.90
11
19.98
10.00
5.76
-467.36
439.88
450.06
425.00
0.27
65.00
800.00
285
16.26
9.51
-2.01
4.90
84.36
-327.02
411.30
418.81
403.50
0.27
250.00
725.00
16
20.53
8.03
21.21
10.00
4.61
-405.00
439.88
448.91
425.00
0.27
65.00
800.00
245
16.19
9.73
22.02
77.21
9.39
-467.36
312.63
482.63
475.00
0.30
100.00
700.00
24
14.13
8.29
12
25.55
13.68
3.27
-195.13
246.11
446.51
420.00
0.26
210.00
725.00
57
22.12
9.42
-25.31
-11.76
2.67
-98.73
34.55
308.33
275.00
0.26
250.00
400.00
3
96.00
5.73
34.36
13.68
1.97
-81.41
221.71
448.62
410.00
0.26
210.00
725.00
47
17.92
9.62
-11.80
24.09
-13.43
-195.13
246.11
491.57
480.00
0.18
395.00
675.00
7
18.71
9.67
13
14
201.28
209.41
1.36
-496.84
689.67
623.75
580.00
0.40
150.00
1250.00
12
29.98
8.76
198.05
195.72
0.02
195.72
202.71
550.00
550.00
0.00
550.00
550.00
3
50.77
7.70
223.10
262.19
0.60
56.35
360.85
622.50
595.00
0.15
500.00
750.00
6
23.10
8.90
15
322.02
322.02
0.06
308.49
335.56
850.00
850.00
0.00
850.00
850.00
2
7.50
11.45
322.02
322.02
0.06
308.49
335.56
850.00
850.00
0.00
850.00
850.00
2
7.50
11.45
160.88
289.81
3.75
-496.84
689.67
700.00
700.00
0.79
150.00
1250.00
3
22.93
9.53
16
76
Impact of Multiple CDO Ratings on Credit Spreads
Table 5-4: Credit Spread of CDO Tranches (Multiple Rating and Rating Code)
7.77
DĂƚƵƌŝƚLJŝŶzĞĂƌƐ;DĞĂŶͿ
485.00
Maximum
89.92
7.97
sŽůƵŵĞŝŶŵh^;DĞĂŶͿ
DĂƚƵƌŝƚLJŝŶzĞĂƌƐ;DĞĂŶͿ
514.79
7.95
Volume in mU^;DĞĂŶͿ
DĂƚƵƌŝƚLJŝŶzĞĂƌƐ;DĞĂŶͿ
131.60
689.67
† scaled by absolute mean values of tranche volumes.
Maximum
3.87
-40.54
4.43
Standard Deviation†
0.79
-498.36
4.20
Median
Minimum
15.63
Mean
4.42
180.00
1,250.00
ĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
Maximum
0.00
0.00
Minimum
0.45
33.00
37.25
1.09
69.00
Median
7.29
202.00
1621
131.60
-40.54
5.00
0.05
Standard Deviation†
134.92
Mean
hŶĂĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
4874
86.14
# of tranches
S&P
Maximum
4.91
Standard Deviation†
-498.36
3.71
Median
Minimum
14.06
Mean
3.09
180.00
900.00
ĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
Maximum
0.00
0.00
Minimum
0.41
33.00
35.67
1.09
65.00
Median
7.26
220.83
1,457
87.60
-40.54
2.43
3.46
Standard Deviation†
132.07
Mean
hŶĂĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
4,596
# of tranches
Moody's
-345.32
Minimum
42.22
2.52
Median
Standard Deviation†
1.64
Mean
7.33
130.00
750.00
ĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
Maximum
0.00
0.00
Minimum
0.49
35.00
39.41
0.99
73.00
Median
7.16
246.69
414
1
Standard Deviation†
122.99
Mean
hŶĂĚũƵƐƚĞĚƌĞĚŝƚ^ƉƌĞĂĚ;ŝŶďƉƐͿ
100.21
1281
Total
sŽůƵŵĞŝŶŵh^;DĞĂŶͿ
# of tranches
Fitch
Code
110.27
-12.54
1.32
21.13
20.59
145.00
15.00
0.49
55.00
53.25
6.59
40.63
22
81.47
-15.01
1.96
5.09
10.71
135.00
19.00
0.41
42.00
49.05
8.52
44.31
86
110.27
-10.56
1.35
12.97
27.59
145.00
20.00
0.60
55.00
60.65
6.49
42.68
17
2
165.34
-56.85
2.95
2.08
7.34
220.00
14.00
0.39
53.00
58.20
8.03
41.86
841
98.24
-56.85
3.98
0.63
4.46
160.00
15.00
0.35
52.00
55.58
8.16
42.83
783
119.97
-50.26
2.48
6.24
9.45
170.00
15.00
0.40
58.00
62.55
7.82
46.10
201
3
104.63
-41.14
8.21
-0.81
3.33
165.00
18.00
0.37
63.00
69.72
6.91
26.40
78
79.59
-41.14
3.05
3.89
8.64
150.00
18.00
0.34
65.00
73.56
6.67
21.52
72
79.59
-36.54
14.38
2.90
2.02
150.00
27.00
0.42
65.00
73.19
7.33
16.68
16
4
84.27
-144.01
-2.76
-27.45
-17.41
225.00
22.00
0.49
70.00
84.87
7.94
32.11
52
122.44
-144.01
23.91
-12.16
-2.17
225.00
24.00
0.53
75.00
95.32
7.70
28.38
74
85.27
-144.01
1.39
-43.43
-31.29
180.00
29.00
0.47
70.00
77.66
8.11
29.06
35
5
240.76
-205.00
2.76
6.93
15.30
350.00
27.00
0.43
100.00
111.66
8.35
28.84
716
240.76
-205.00
2.65
6.93
15.36
350.00
20.00
0.44
95.00
110.28
8.40
28.99
686
175.20
-156.62
10.80
3.68
4.46
260.00
24.00
0.44
100.00
110.07
7.87
30.77
168
6
292.55
-216.62
3.93
15.11
16.65
500.00
29.00
0.48
135.00
137.65
8.19
22.59
159
292.55
-216.62
9.30
-0.82
7.18
500.00
30.00
0.47
130.00
135.91
8.28
27.50
156
129.69
-167.37
4.52
-20.43
-11.89
265.00
37.00
0.33
125.00
127.08
8.80
27.53
76
7
325.84
-310.95
20.42
-6.33
6.06
525.00
53.00
0.56
147.50
198.30
7.87
22.26
66
325.84
-310.95
16.99
-13.42
7.94
525.00
95.00
0.52
150.00
207.28
7.93
20.66
58
138.67
-246.79
2.05
-29.66
-33.42
325.00
55.00
0.44
133.00
153.62
8.37
27.47
39
8
532.87
-345.32
2.35
21.89
37.80
800.00
16.00
0.40
242.50
247.80
8.16
22.52
738
490.00
-345.32
2.09
22.64
40.60
750.00
16.00
0.40
245.00
248.37
8.13
22.22
670
485.00
-345.32
9.04
21.10
11.42
750.00
53.00
0.38
250.00
243.09
7.81
21.52
204
9
419.40
-498.36
-6.32
-54.35
-21.57
800.00
45.00
0.52
202.50
260.04
8.54
20.72
180
419.40
-498.36
3.81
-58.51
-35.80
800.00
45.00
0.55
190.00
246.64
8.50
22.25
190
108.33
-343.05
0.78
-123.40
-114.79
350.00
61.00
0.28
210.00
214.33
8.33
18.64
43
10
493.31
-386.88
0.85
251.99
204.14
900.00
100.00
0.26
625.00
572.76
8.25
11.22
70
514.79
-386.88
0.67
261.93
237.39
900.00
90.00
0.25
625.00
591.67
6.72
8.98
63
372.94
-197.20
1.90
75.17
90.60
650.00
250.00
0.32
500.00
457.76
13.03
14.58
17
11
439.88
-467.36
5.25
10.00
20.92
800.00
100.00
0.27
425.00
452.13
9.56
16.37
270
411.30
-405.00
3.92
10.00
25.09
800.00
170.00
0.26
425.00
451.50
9.82
16.71
236
411.30
-327.02
64.22
12.22
2.42
750.00
170.00
0.29
450.00
455.83
8.25
17.25
40
12
246.11
-174.53
2.25
14.40
32.21
725.00
250.00
0.23
410.00
446.58
9.76
20.83
48
221.71
-338.30
57.81
-5.67
-1.95
700.00
210.00
0.26
400.00
417.05
9.14
21.52
55
34.55
-195.13
0.96
-106.57
-91.36
475.00
65.00
0.48
337.50
317.50
7.40
54.42
6
13
360.85
-496.84
1.68
209.41
163.85
208.55
201.69
0.02
201.69
203.98
550.00
550.00
0.00
550.00
550.00
7.70
50.77
3
14
689.67
56.35
0.65
216.11
264.75
1250.00
500.00
0.31
590.00
666.82
8.59
28.61
11
335.56
308.49
0.06
322.03
322.03
750.00
150.00
0.32
560.00
538.75
8.76
38.79
8
68.78
56.35
0.14
62.57
62.57
725.00
500.00
0.26
612.50
612.50
8.50
12.80
2
15
335.56
308.49
0.06
322.03
322.03
850.00
850.00
0.00
850.00
850.00
11.45
7.50
2
850.00
850.00
0.00
850.00
850.00
11.45
7.50
2
16
Impact of Multiple CDO Ratings on Credit Spreads
77
Table 5-5: Credit Spread of CDO Tranches (Rating Agency and Rating Code)
78
Impact of Multiple CDO Ratings on Credit Spreads
5.5.2 Impact of Multiple Ratings
In Table 5-4 we compare in detail the credit spreads of single, double and triple ratings. This allows us to perform first tests for Hypothesis 1. With mean values of
170.35 bps for unadjusted credit spreads (22.36 bps for adjusted credit spreads) for
single ratings, 136.83 bps (16.23 bps) for double ratings and 134.83 bps (14.04 bps)
for triple ratings we observe that with an increasing number of ratings the level of both
unadjusted and adjusted credit spread decreases. In addition to the negative correlation
between number of outstanding ratings and credit spreads we also observe a positive
relationship between number of tranche ratings and average tranche size as the latter
increases from 45.2 mUSD (single ratings) to 115.1 mUSD (triple ratings). This finding seems to be reasonable since an increasing tranche size allows allocation of rating
fees to a broader capital basis.
Table 5-6: Robustness Checks for the Grouping Factor Multiple Ratings
ANOVA
Volume
Maturity
Unadjusted Credit Spread
Adjusted Credit Spread
Between Groups*
Within Groups
Total
Between Groups
Within Groups
Total
Between Groups
Within Groups
Total
Between Groups
Within Groups
Total
Sum of Squares
1109,657.48
203,426,186.18
204,535,843.66
243.25
30,792.20
31,035.46
779,357.33
107,459,348.51
108,238,705.84
205,843.88
24,739,518.43
24,945,362.31
Statistic**
28.94877899
12.61674765
29.62322859
29.22022976
22.05001686
18.38957736
20.21044313
13.32254807
df
2
513
513
Mean Square
554,828.74
39,654.23
F
13.99
Sig.
0.0000
2
513
513
121.63
6.00
20.26
0.0000
2
513
513
389,678.67
20,947.24
18.60
0.0000
2
5130
5132
102,921.94
4,822.52
21.34
0.0000
df1
2
2
2
2
2
2
2
2
df2
778.98
1,020.08
668.65
864.24
630.30
617.24
588.85
562.70
Robust Tests of Equality of Means
Volume
Welch
Brown-Forsythe
Maturity
Welch
Brown-Forsythe
Unadjusted Credit Spread
Welch
Brown-Forsythe
Adjusted Credit Spread
Welch
Brown-Forsythe
* Groups are defined as Single, Double and Triple Ratings
** Asymptotically F distributed
Sig.
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
Impact of Multiple CDO Ratings on Credit Spreads
79
In order to further determine if ‫ ݐ݅ܵܥ‬, ‫ ݐ݅ܵܥܣ‬, maturity and tranche volume are indeed
significantly different for single, double and triple ratings we perform several robustness checks which confirm our findings on a statistically significant level (see Table 56). Relying on a one-way ANOVA as well as robust tests of equality (Welch and
Brown-Forsythe) all four tranche characteristics are statically significantly different
from each other. In addition, we perform a test of homogeneity of variances which
leads to the very same results. Thus, with regard to Hypothesis 1 we are able reject the
null hypothesis, and not reject the alternative hypothesis respectively, allowing us to
argue that there is a negative correlation between the number of outstanding ratings
and the level of credit spreads. Each additional rating leads to a lower credit spread –
both for adjusted and unadjusted credit spreads.
5.5.3 Decreasing Reduction of Underlying Tranche Spreads
In the following, we analyze if the level of correlation between credit spreads and
number of ratings changes when moving from single to double or from double to triple
ratings. In Table 5-7 we compare in detail the mean differences of unadjusted and adjusted credit spreads between single, double and triple ratings. In addition, we display
mean differences for maturity and tranche volume. Due to different interest term structure tranche maturity is by definition an important factor when discussing attributes of
debt instruments. Tranche volume is especially relevant for structured credit transactions as rating costs are usually fixed cost decrease per investor when tranches are of
significant seize. Furthermore, small tranches are usually less transparent than larger
ones as they are sometimes tailor-made for large investors or sold in club markets.
Therefore, tranche volume is an important factor.
According to Table 5-4 tranche volume seems to positively correlate with the number
of outstanding ratings. Thus, it is reasonable to analyze incremental correlation between these two factors if moving from single to double or double to triple ratings.
While average maturity appears to increase from single to double ratings and to decrease from double to triple ratings (see Table 5-4), no clear pattern can be derived for
tranche maturities. The average tranche volume increases from single to double and
double to triple ratings with decreasing marginal differences: Volume grows by 38.61
80
Impact of Multiple CDO Ratings on Credit Spreads
mUSD on average from single to double ratings, whereas growth from double to triple
ratings amounts to 31.28 mUSD on average. With 33.51 bps (single to double ratings)
versus 25.60 bps (double to triple ratings) this pattern also holds for unadjusted credit
spreads. Thus, the reduction of unadjusted credit spread levels decreases with an increase in the number of outstanding ratings. In contrast, exclusion of systematic credit
risk (adjusted credit spreads) leads to a different result: reduction of credit spreads levels increases from 6.13 bps between single and double ratings to 17.07 bps between
double and triple ratings.
Since we already proved that variances of the variables volume, maturity and
(un)adjusted credit spreads are unequal (see Table 5-6), we are able to apply a GamesHowell test in order to check for significance of mean differences on the 0.05 significance level. Mainly, we are able to confirm significance at a 0.05 level. Cross-checking
the Games-Howell results with the testing algorithms of Tamhane’s T2, Dunnett’s T3
as well as Dunnett’s C leads to the same significance levels.
Regarding Hypothesis 2 it becomes clear that in the case of unadjusted credit spreads
we are not able to reject the null hypothesis but the alternative hypothesis. For the adjusted credit spreads we note in turn a rejection of the null hypothesis and a nonrejection of the alternative hypothesis. We could not find empirical support on all levels for the hypothesis stating that marginal tranche spread reduction decreases when
adding additional rating agencies. Against the background of these results we are not
able to observe a clear pattern relating to the question whether marginal ratings lead to
increasing, decreasing or constant credit spreads. However, since we created the adjusted credit spreads in order to analyze the idiosyncratic credit risk without any dilution of the systematic credit risk, multiple ratings should have the highest impact on
adjusted credit spreads. Thus, we analyze mean spread differences of adjusted credit
spreads and interpret increasing credit spread reduction as representing a selection bias. Also missing levels of significance for mean difference between single and double
ratings (adjusted credit spreads) does not change this view since mean difference between double and triple ratings are significant (Table 5-7).
Impact of Multiple CDO Ratings on Credit Spreads
81
Table 5-7: Multiple comparisons of underlying tranche spread differences
Dependent Variable
Volume (in mUSD)
(I) MultipleRatings
Single Rating
Double Rating
Triple Rating
Maturity (in Years)
Single Rating
Double Rating
Triple Rating
Unadjusted Credit Spread (in bps)
Single Rating
Double Rating
Triple Rating
Adjusted Credit Spread (in bps)
Single Rating
Double Rating
Triple Rating
(J) MultipleRatings Mean Difference (I-J)
Double Rating
-38.6083*
Triple Rating
-69.8854*
Single Rating
38.6083*
Triple Rating
-31.2771*
Single Rating
69.8854*
Double Rating
31.2771*
Double Rating
-0.5915*
Triple Rating
-0.0656
Single Rating
0.5915*
Triple Rating
0.5259*
Single Rating
0.0656
Double Rating
-0.5259*
Double Rating
33.5144*
Triple Rating
59.1163*
Single Rating
-33.5144*
Triple Rating
25.6019*
Single Rating
-59.1163*
Double Rating
-25.6019*
Double Rating
6.1288
Triple Rating
23.1991*
Single Rating
-6.1288
Triple Rating
17.0704*
Single Rating
-23.1991*
Double Rating
-17.0704*
Std. Error
5.7876
11.8965
5.7876
11.1395
11.8965
11.1395
0.1365
0.1452
0.1365
0.0755
0.1452
0.0755
10.5623
11.0834
10.5623
4.6885
11.0834
4.6885
6.4510
6.8556
6.4510
2.7441
6.8556
2.7441
Sig.
0.0000
0.0000
0.0000
0.0141
0.0000
0.0141
0.0001
0.8937
0.0001
0.0000
0.8937
0.0000
0.0047
0.0000
0.0047
0.0000
0.0000
0.0000
0.6091
0.0023
0.6091
0.0000
0.0023
0.0000
* The mean difference is significant at the .05 level.
5.5.4 CDO Tranches rated by Fitch
Sorting the CDO tranches for rating agencies result in 3 different groups (Table 5-5):
1,281 tranches rated by Fitch, 4,596 tranches rated by Moody’s and 4,874 tranches
rated by S&P adding up to 10,751 cases in total. Since multiple ratings exist, many
tranches are included in more than one group. We now focus on the average rating per
rating agency and compare the rating outcome. Specifically, we test according to Hypothesis 3 if the average Fitch rating is different from the average S&P or Moody’s
rating or not.
Table 5-8 provides a detailed overview of mean ratings. Since numbers of tranches per
agency differ substantially (Fitch 1,281 tranches, Moody’s 4,596 tranches, S&P
4,874), this figure only displays a first illustration of mean ratings. However, with a
mean rating code of 4.8595 Fitch ratings obtain the highest credit quality; Moody’s
and S&P in turn achieve lower mean rating levels. Starting from here, we focus on the
detailed ratings assigned by the agencies for the very same tranche. This approach
leaves us with two different samples: multiple and joint ratings. Multiple ratings come
82
Impact of Multiple CDO Ratings on Credit Spreads
Table 5-8: Comparison of Rating Outcomes (Rating Agencies and Rating Code)
Average Rating Code
Rating Agency
Total
Single
Fitch
Moody's
S&P
Total
Double RatMoody's/
Fitch/ S&P
Moody's/
Triple Ratings
Fitch/ MooJoint Ratings
Moody's/
Fitch/ S&P
Moody's/
# of rated
ŝŶйŽĨ
Tranches Total Sample
5,133
ϭϬϬ͘ϬϬй
Fitch
4.8595
Moody's
4.9238
Average Notch Difference*
S&P
4.9179
Moody's ./.
S&P
0.0059
68
37
166
271
ϭ͘ϯϮй
Ϭ͘ϳϮй
ϯ͘Ϯϯй
ϱ͘Ϯϴй
3,649
303
154
ϳϭ͘Ϭϵй
ϱ͘ϵϬй
ϯ͘ϬϬй
5.1023
5.0260
5.0455
756
ϭϰ͘ϳϯй
4.5132
4.5847
4.5556
0.0291**
4.9219
4.9067
4.7186
0.0152**
4,405
1,059
910
Fitch ./.
S&P
-0.0584
Moody's ./.
Fitch
0.0644
7.2500
4.6486
4.8373
4.9918
4.6818
4.6000
4.6626
4.9794
5.1254
0.0123**
-0.0231
0.0195
-0.0423**
0.0714**
-0.0368**
0.0626**
* In case of (-Ϳ͕ƐƵďƚƌĂŚĞŶĚŚĂƐƌĂƚĞĚŽŶĂǀĞƌĂŐĞůŽǁĞƌ͘
**Significant at the 0.05 level.
as single, double and triple ratings; joint ratings only pair wise. Single ratings are not
directly applicable to compare different rating outcomes; however, they give an indication if a specific rating agency is favored by issuers for a specific part of the seniority
structure.
With a mean value of 7.25 single ratings by Fitch correspond to a significantly lower
credit quality than single ratings by Moody’s (4.6486) and S&P (4.8373). This indicates that Fitch rates more junior tranches than the other agencies do. What is even
more revealing is the analysis of double and triple ratings. Based on notch differences
we observe that Fitch – when directly compared with Moody’s and Fitch –on average
assigns a better rating as do Moody’s and S&P for the very same tranche in double ratings (Table 5-8). In turn, S&P ratings document on average a higher credit quality as
do Moody’s ratings for the very same tranches. With 4.5132 (Fitch), 4.5556 (S&P) and
4.5847 (Moody’s), this pattern also holds for triple ratings. In cases where a tranche is
rated by all three rating agencies, Fitch ratings are on average better (e.g. lower in
terms of rating codes) than the corresponding ratings by Moody’s and S&P. Moody’s
in turn assigns the highest rating codes (lowest credit quality). Thus, the biggest notch
difference is between Moody’s and Fitch. The analysis of jointly-rated tranches sup-
Impact of Multiple CDO Ratings on Credit Spreads
83
ports these results, since Fitch again assigns on average the best rating and the lowest
rating code in direct comparison to Moody’s and S&P.
For triple and joint ratings all notch differences are significant on the 0.05 level (see
Table 5-8). For double ratings significance of mean notch differences is only given for
the combination Moody’s/S&P. However, it needs to be considered that the pairs
Fitch/S&P and Moody’s/Fitch are comparatively rare with 303 and 154 tranches. Significance on the level of jointly-rated tranches for Fitch/S&P as well as Moody’s/Fitch
confirms the results obtained throughout the analysis of triple ratings on a larger scale
(1,059 and 910 tranches). With regard to Hypothesis 3 we can reject the null hypothesis and not reject the alternative hypothesis. This means that the average Fitch rating is
different and significantly better than the corresponding Moody’s and S&P ratings.
5.5.5 Regression Analysis
To assess the impact various factors have on the underlying credit spread the application of a regression analysis is a widely accepted measure in financial literature and
also commonly used in the context of multiple ratings (e.g. Kish et al., 1999; Jewell
and Livingston, 1999; Vink and Thibeault, 2008). So far we limited our analysis to univariate statistics. In the following we perform a series of regression analyses in order
to specify the impact multiple ratings have on the underlying tranche credit spread. We
define the unadjusted credit spread ‫ ݐ݅ܵܥ‬and the adjusted credit spread ‫ ݐ݅ܵܥܣ‬as the dependent variables and multiple ratings as well as several other factors (e.g. volume) as
independent variables and perform a regression analysis according to Vink and Thibeault (2008). We define our valuation model as follows:
‫ ߙ = ݐ݅ܵܥ‬+ ߚ1 ‫ ݅ݏ݃݊݅ݐܴܽ ݈݁݌݅ݐ݈ݑܯ‬+ ߚ2 ܴܽ‫ ݅݁݀݋ܥ ݃݊݅ݐ‬+ ߚ3 ܶ‫݅݁݉ݑ݈݋ܸ ݄݁ܿ݊ܽݎ‬
+ ߚ4 ‫ ݅ݕݐ݅ݎݑݐܽܯ‬+ ߚ5 ܶ‫ ݅݁݌ݕܶ ݊݋݅ݐܿܽݏ݊ܽݎ‬+ ߚ6 ‫݅ݕܿ݊݁ݎݎݑܥ‬
+ ߚ7 ܻ݁ܽ‫ ݅݁ܿ݊ܽݑݏݏܫ ݂݋ ݎ‬+ ߝ݅
with:
Multiple Ratingsi :
Zero-one variables for multiple ratings (single and double ratings); takes one if the tranche received a corresponding multiple
84
Impact of Multiple CDO Ratings on Credit Spreads
rating, zero if otherwise; triple ratings function as a base case
and are thus excluded from the analysis
Rating Codei :
Zero-one variables for average rating codes (1, 2, 3, 4, 5, 6, 7, 8,
9, 10, 11, 12, 13, 15); takes one if the tranche received a rating
of the corresponding rating code, zero if otherwise; rating code
16 functions as a base case and is thus excluded from the analysis, rating code 14 is also not displayed since no data points are
existing for this rating code.
In a second set of regression analyses (Set B) we replace zeroone variables with a single metric scale representing the tranche
average rating code (Mean Rating Code).
Tranche Volumei
Volume of tranche i in mUSD
Maturityi
Maturity of tranche i in years
Transaction Typei
Zero-one variables for transaction type (CBO and CLO); takes
one if the transaction type corresponds to the specific type, zero
if otherwise; transaction type Exotic functions as a base case
and is thus excluded from the analysis
Currencyi :
Zero-one variable for currency (USD and other currencies);
takes one if the transaction is denominated in USD, zero if otherwise
Year of Issuancei :
Zero-one variables for year of issuance (2005, 2006, 2007);
takes one if the tranche was issued in the corresponding year,
zero if otherwise; year 2004 functions as a base case and is thus
excluded from the analysis
We explicitly include multiple ratings as two nominal zero-variables (single and
double ratings) and not as a scale metric (number of outstanding ratings) as applied by
Vink and Thibeault (2008). This differentiation allows us to isolate the impact of the
individual characteristic (single or double ratings). Different rating codes like zero-one
Impact of Multiple CDO Ratings on Credit Spreads
85
variables consider the seniority structure of a CDO transaction and test for increasing
spread levels with decreasing credit quality. Maturity and tranche volume are a natural
choice to be incorporated in our valuation model since both variables are crucial for
each CDO tranche. Literature (Vink and Thibeault, 2008) and the comparison of different CDO credit spread indices in our data set already reveal that credit spreads do
not necessarily have the same level for different transaction types. Specifically, exotic
CDOs are traded at higher spreads compared to CLOs and CBOs.
The majority of tranches is denominated in USD (81.2%) followed by EUR. Besides,
with approx. 15 tranches other currencies play no meaningful role and we thus merged
EUR and other currencies into one group. Relevance of currency as a major impact
variable relates to the fact that rating agencies assign to different recovery rates U.S,
and European CDOs. Recovery rates in turn directly impact losses to which credit
spreads are positively correlated. Mostly the collateral pool consists of assets located
in the transaction’s currency area (e.g. U.S. assets are typically used as a collateral
pool of transactions denominated in USD). Thus, there exists a link between the transaction’s currency and the underlying tranche spread. Finally, we also included the year
of issuance into our valuation model. Year of issuance is of particular interest since we
observe over our whole sample period (2004-2007) decreasing credit spreads on an
overall basis. Specifically, in case of unadjusted credit spreads we expect high significance levels. Prior to our regression analysis we controlled for normal distribution of
all independent variables (KS-test).
We display the results of our regression analysis in Table 5-9. As outlined before, we
perform our analysis for two dependent variables: unadjusted and adjusted credit
spreads. In addition, we analyze two sets of independent factors (Set A and B). We
created two compositions of independent variables because regression analysis of Set
A leaves us with very modest signs of multicollinearity for selected zero-one variables
of the average rating code (e.g. 3, 9, 12). Since we do not want to dilute our analysis or
our R2 through imprecise data, we replaced the zero-one variables by the scale metric
mean rating code of each tranche in a second trial (Set B). However, we kept the zeroone variables for Set A due to information about seniority structure of a CDO.
86
Impact of Multiple CDO Ratings on Credit Spreads
In addition we also checked for outliers. Since the exclusion of outliers impacted the
final results only on a very modest basis (e.g. R2 of regression analysis Set B with unadjusted credit spread only increased from 0.7080 to 0.7250) we decided to apply the
regression analysis of Table 5-9 to the original data set of 5.133 different CDO
tranches. Thus, we maintain data consistency with the prior analysis of our empirical
section.
In each regression (adjusted and unadjusted credit spreads) of Set A and Set B all variables have significant impact on the credit spread level. A large majority of variables
is even significant on the 0.01 level. Concerning multiple ratings we observe that single ratings lead to higher credit spreads than double ratings. This finding supports the
results of our preceding univariate analysis of negative correlation between number of
ratings and level of credit spreads (see Table 5-4). It is particularly interesting that the
impact of standardized coefficients is higher for adjusted credit spreads in both sets.
Thus, the exclusion of the systematic credit risk proves to help to isolate the specific
impact multiple ratings have on credit spread levels. A second indication for this fact
is the lower impact of mean rating code on adjusted credit spreads than on unadjusted
credit spreads in Set B (0.217 versus 0.8560). Based on Hypothesis 1, we can reject
the null hypothesis and not reject the alternative hypothesis. In addition, the results of
our regression analysis emphasize a negative correlation between number of ratings
and credit spread. Thus, it confirms our results illustrated on a univariate basis in Table
5-4.
The coefficients of different zero-variables relating to the rating codes decrease with
increasing rating codes, which equals a higher spread level for decreasing credit quality and thus confirms the results of Vink and Thiebeault (2008). Relating to the standardized coefficients of the regression analysis in Set A as well as in Set B zero-one
variables of the average rating code have on average the highest impact. Again, this
finding is not surprising and confirmed in Set B through the high impact of mean rating code. In contrast to our univariate analysis (Table 5-4) the impact of tranche volume is negligible in both sets and around zero in absolute terms.
A possible explanation could be the mode of payment for CDO ratings. The issuer
pays a fee, which is predominately a percentage of the underlying tranche volume (e.g.
Impact of Multiple CDO Ratings on Credit Spreads
87
4.5 bps), which dilutes the incentive to rate only tranches with large volumes by multiple rating agencies. Maturity in turn has a slightly higher impact and can be explained by different interest rate term structures. For the transaction structure we observe that CBO structures lead to lower credit spreads as documented for CLOs. A potential explanation for this finding could be the fact that the collateral pool of a CBO is
expected to be more liquid than the collateral pool of a CLO. Thus, the whole transaction becomes more price sensitive from the investor’s angle. The results relating to the
zero-one variables of the issuance year are in line with the development of CDS
spreads as a benchmark over the same period. Thus, this dummy variable behaves as
expected. Specifically, the comparably low credit spreads documented for 2007 result
in a decreasing unstandardized coefficient of the independent variable “YEAR OF ISSUANCE 2007”.
Since our valuation model (Table 5-9) consists of a number of different independent
pricing factors, we additionally perform several robustness checks in order to control
for potential effects left undetected throughout our initial regression analysis (Set A &
B). We want to control for the explanatory power of multiple ratings on credit spread
levels. First, each independent pricing factor is individually regressed on unadjusted
and adjusted credit spreads. Second, we include the dummy variables relating to multiple ratings (single and double ratings) and again perform a regression analysis for
each independent pricing factor. If multiple ratings indeed incorporate explanatory
power the generated R2, values should increase following the inclusion of multiple rating dummy variables. In Table 5-10 we have exemplarily outlined the results of the
robustness checks for the pricing factor tranche volume. Throughout the analysis of
credit spreads we observed a positive correlation between number of outstanding ratings and tranche volume. Thus, the performed robustness checks should give us an insight, if we have covered a mere size effect throughout our valuation model. For both
dependent variables we note increasing R2 values as well as high levels of significance
for the two dummy variables of multiple ratings. Even though the increase is rather
modest on an absolute basis, relative differences are high. The pattern documented for
tranche volume (Table 5-10) also holds for the other independent pricing factors.
936.4885***
44.8969ΎΎΎ;Ϭ͘ϬϲϵϭͿ
26.4475ΎΎΎ;Ϭ͘ϬϳϮϵͿ
-845.4537***(-Ϯ͘ϳϬϲϴͿ
-821.8987***(-Ϭ͘ϳϮϲϱͿ
-824.4751***(-Ϯ͘ϭϮϱϯͿ
-824.2198***(-Ϭ͘ϳϮϬϭͿ
-795.7164***(-Ϭ͘ϱϲϵϮͿ
-767.0254***(-ϭ͘ϴϴϰϭͿ
-757.2764***(-Ϭ͘ϵϴϵϱͿ
-697.6798***(-Ϭ͘ϱϮϰϴͿ
-636.7029***(-ϭ͘ϱϲϴϮͿ
-624.0415***(-Ϭ͘ϴϰϭϱͿ
-308.5228***(-Ϭ͘ϮϯϱϴͿ
-412.2900***(-Ϭ͘ϲϱϬϮͿ
-402.7010***(-Ϭ͘ϮϵϬϲͿ
-234.6547***(-Ϭ͘ϬϳϴϬͿ
-0.0132***(-Ϭ͘ϬϭϴϭͿ
-3.5378***(-Ϭ͘ϬϱϵϵͿ
-54.3929***(-Ϭ͘ϬϮϮϳͿ
-42.2555***(-Ϭ͘ϭϰϱϯͿ
-17.2768***(-Ϭ͘ϬϰϲϱͿ
-30.0737***(-Ϭ͘ϬϵϬϭͿ
-34.6765***(-Ϭ͘ϭϭϵϰͿ
-16.5672***(-Ϭ͘ϬϯϮϯͿ
Coefficients
5,133
1,074.84
0.8350
* Significance at the 0.1 level
** Significance at the 0.05 level
*** Significance at the 0.01 level
Values in brackets refer to standardized coefficients.
(ŽŶƐƚĂŶƚͿ
SINGLERATING
DOUBLERATING
RATINGCODE 1
RATINGCODE 2
RATINGCODE 3
RATINGCODE 4
RATINGCODE 5
RATINGCODE 6
RATINGCODE 7
RATINGCODE 8
RATINGCODE 9
RATINGCODE 10
RATINGCODE 11
RATINGCODE 12
RATINGCODE 13
RATINGCODE 15
TRANCHE VOLUME
MATURITY
TRANSACTIONTYPE CBO
TRANSACTIONTYPE CLO
CURRENCY
YEAR 2005
YEAR 2006
YEAR 2007
Independet Variables
N
F
2
R
407.6073***
Ϯϱ͘ϲϴϬϵΎΎΎ;Ϭ͘ϬϴϮϰͿ
ϮϮ͘ϱϴϭϵΎΎΎ;Ϭ͘ϭϮϵϲͿ
-346.2465***(-Ϯ͘ϯϬϵϭͿ
-332.1769***(-Ϭ͘ϲϭϭϲͿ
-340.0975***(-ϭ͘ϴϮϲϮͿ
-350.9642***(-Ϭ͘ϲϯϴϳͿ
-356.1795***(-Ϭ͘ϱϯϬϳͿ
-329.7169***(-ϭ͘ϲϴϳϭͿ
-335.3720***(-Ϭ͘ϵϭϮϵͿ
-348.7539***(-Ϭ͘ϱϰϲϱͿ
-311.8145***(-ϭ͘ϱϵϵϳͿ
-371.6635***(-ϭ͘ϬϰϰϬͿ
-133.8774***(-Ϭ͘ϮϭϯϭͿ
-314.1313***(-ϭ͘ϬϯϭϵͿ
-301.9321***(-Ϭ͘ϰϱϯϵͿ
-129.7039***(-Ϭ͘ϬϴϵϵͿ
-0.01733***(-Ϭ͘ϬϰϵϲͿ
-8.7632***(-Ϭ͘ϯϬϵϭͿ
-49.6281***(-Ϭ͘ϬϰϯϮͿ
-3.6198*(-Ϭ͘ϬϮϱϵͿ
-13.6782***(-Ϭ͘ϬϳϲϳͿ
-4.0772(-Ϭ͘ϬϮϱϰͿ
-5.7583**(-Ϭ͘ϬϰϭϯͿ
-9.3106***(-Ϭ͘ϬϯϳϴͿ
Coefficients
5,133
82.24
0.2790
;ŽŶƐƚĂŶƚͿ
SINGLERATING
DOUBLERATING
MEANRATINGCODE
TRANCHE VOLUME
MATURITY
TRANSACTIONTYPE CBO
TRANSACTIONTYPE CLO
CURRENCY
YEAR 2005
YEAR 2006
YEAR 2007
Independet Variables
N
F
2
R
13.1810**
ϱϮ͘ϬϰϲϯΎΎΎ;Ϭ͘ϬϴϬϮͿ
28.01530***(0.ϬϳϳϮͿ
ϯϯ͘ϯϳϰϯΎΎΎ;Ϭ͘ϴϱϲϭͿ
Ϭ͘ϬϯϮϯΎΎΎ;Ϭ͘ϬϰϰϰͿ
-3.6400***(-Ϭ͘ϬϲϭϲͿ
-56.6195***(-Ϭ͘ϬϮϯϳͿ
-34.4079***(-Ϭ͘ϭϭϴϯͿ
-16.3814***(-Ϭ͘ϬϰϰϭͿ
-29.2951***(-Ϭ͘ϬϴϳϴͿ
-30.6680***(-Ϭ͘ϭϬϱϲͿ
-12.7319***(-Ϭ͘ϬϮϰϴͿ
Coefficients
5,133
1,127.36
0.7080
Dependent Variable: Unadjusted Credit Spread
Dependent Variable: Unadjusted Credit Spread
Dependent Variable: Adjusted Credit Spread
SET B
SET A
54.5407***
Ϯϵ͘ϮϵϳϯΎΎΎ;Ϭ͘ϬϵϰϬͿ
Ϯϯ͘ϳϴϮϭΎΎΎ;Ϭ͘ϭϯϲϱͿ
ϰ͘ϬϲϳϳΎΎΎ;Ϭ͘ϮϭϳϯͿ
-0.0135***(-Ϭ͘ϬϯϴϳͿ
-8.8089***(-Ϭ͘ϯϭϬϳͿ
-54.5171***(-Ϭ͘ϬϰϳϱͿ
-5.5371***(-Ϭ͘ϬϯϵϳͿ
-16.6892***(-Ϭ͘ϬϵϯϱͿ
-4.0414(-Ϭ͘ϬϮϱϮͿ
-4.6811*(-Ϭ͘ϬϯϯϲͿ
-8.3343**(-Ϭ͘ϬϯϯϴͿ
Coefficients
5,133
79.28
0.1460
Dependent Variable: Adjusted Credit Spread
88
Impact of Multiple CDO Ratings on Credit Spreads
Table 5-9: Impact of Multiple Ratings (Multiple Regression Analysis)
-0.1723***(-0.2368Ϳ
-0.1759***(-0.2418Ϳ
* Significance at the 0.1 level
** Significance at the 0.05 level
*** Significance at the 0.01 level
Values in brackets refer to standardized coefficients.
131.0579***
47.0774***(0.0725Ϳ
20.2139***(0.0557Ϳ
150.0248***
;ŽŶƐƚĂŶƚͿ
SINGLERATING
DOUBLERATING
RATINGCODE 1
RATINGCODE 2
RATINGCODE 3
RATINGCODE 4
RATINGCODE 5
RATINGCODE 6
RATINGCODE 7
RATINGCODE 8
RATINGCODE 9
RATINGCODE 10
RATINGCODE 11
RATINGCODE 12
RATINGCODE 13
RATINGCODE 15
TRANCHE VOLUME
MATURITY
TRANSACTIONTYPE CBO
TRANSACTIONTYPE CLO
CURRENCY
YEAR 2005
YEAR 2006
YEAR 2007
Coefficients
Coefficients
5,133
114.90
0.0630
Independet Variables
5,133
318.54
0.0585
;ŽŶƐƚĂŶƚͿ
SINGLERATING
DOUBLERATING
MEANRATINGCODE
TRANCHE VOLUME
MATURITY
TRANSACTIONTYPE CBO
TRANSACTIONTYPE CLO
CURRENCY
YEAR 2005
YEAR 2006
YEAR 2007
Independet Variables
N
F
2
R
5,133
32.16
0.0062
2.0978
21.4132***(0.0ϲϴϳͿ
16.2711***(0.0937Ϳ
-0.0256***(-0.0732Ϳ
0.0323***(0.0789Ϳ
Coefficients
5,133
23.53
0.0136
16.4172***
Coefficients
Dependent Variable: Adjusted Credit Spread
Dependent Variable: Unadjusted Credit Spread
N
F
2
R
Regression Analysis
Regression Analysis
Impact of Multiple CDO Ratings on Credit Spreads
89
Table 5-10: Robustness Checks (Controlling for Size Effect)
90
Impact of Multiple CDO Ratings on Credit Spreads
5.6 Conclusion
The main objective of this paper is to analyze the impact of multiple CDO ratings on
credit spreads of the respective tranches. The analysis is performed on a data set of
more than 5,000 CDO tranches for which we calculated index-adjusted credit spreads
by subtracting an average CDO Credit Spread Index ‫ ܿݎݐ݆ݔ݁݀݊ܫ ܵܥ‬from the individual
unadjusted credit spread ‫ ݐ݅ܵܥ‬to isolate the specific credit risk per CDO tranche. Thereby, we are able to analyze idiosyncratic credit spreads of different tranches without
any dilution from systematic credit risk. We argue that each additional rating incorporates new incremental information and thus reduces information asymmetry between
the issuer and the investor. Reduced information asymmetry increases transparency,
thereby lowers investors’ demand for risk premiums and leads to lower credit spreads.
The motivation for this empirical analysis becomes especially relevant when considering the current financial crisis. Among others, information asymmetries between issuers and investors and misaligned incentive structures for issuers along the structuring
process of CDOs lead to a situation where only insufficient information was shared
with investors.
Our key findings are threefold: First, we find that on average credit spreads indeed decrease with an increasing number of ratings. The obtained negative correlation between multiple ratings and adjusted credit spreads is statistically robust and crosschecked for various factors. In addition, we developed a valuation model incorporating
multiple ratings as an independent variable. In a regression of the number of outstanding ratings on credit spreads controlling for various factors (e.g. maturity), we document significant impact levels for multiple ratings. We show that in addition to other
pricing factors (e.g. credit quality) the number of outstanding ratings incorporates explanatory power with respect to the pricing structure of CDO credit spreads. These results empirically support our argumentation stating that additional ratings reduce existing information asymmetries between issuer and investors and thus lower credit spread
premiums demanded by investors. Introduction of index-adjusted credit spreads reduced the impact of variables linked to the tranche’s credit quality in our valuation
model and in turn led to a further increase of observed influence levels for multiple
ratings.
Impact of Multiple CDO Ratings on Credit Spreads
91
Second, even with decreasing spread levels in place, we were not able to confirm the
hypothesis that marginal tranche spread reduction decreases with the number of published ratings. These results make it rather difficult to determine and recommend an
optimal number of ratings an investor should opt for when structuring a CDO. Additional ratings always come with additional costs; thus, the incremental value of additional ratings through spread reduction should at least amount to the level of costs associated with an additional rating. CDO rating costs are viewed to be in the range of
4.5 bps of the underlying tranche volume. However, in both cases (single to double
and double to triple ratings) documented spread reduction is a lot higher. Therefore,
investors have an economic incentive to seek multiple ratings.
Third, we reviewed the outcome of Fitch ratings in direct comparison to Moody’s and
S&P ratings. Research exists targeting this issue from the perspective of corporate
bonds (e.g. Jewell and Livingston, 1999) while, again, the role of Fitch ratings within
structured finance transactions has not been analyzed before. We found that in the case
of joint (pair wise) ratings, on average Fitch assigned a higher credit quality (e.g. better rating) than its competitors Moody’s and S&P did for the very same CDO tranche.
Since Fitch is by far the smallest of the three rating agencies offering services in the
field of CDO ratings, we see a potential explanation in the form of a selection bias. Issuers only assign a CDO rating to Fitch if the expected outcome is better than the one
obtained by Moody’s or S&P.
The role of multiple ratings in valuation models for corporate bonds has been widely
discussed in the literature. However, the transfer of the results to the field of securitization is rather inappropriate since the CDO rating process is substantially different from
the rating process of corporate bonds: CDO ratings are solicited by the issuer, who
chooses the rating agencies and controls the rating process. Additionally, issuer and
rating agency are in close contact throughout the rating process – a behavior heavily
criticized in the recent past by politicians and regulatory authorities. This negotiation
process as well as the role of rating agencies in structured finance business and their
business models will be intensely discussed in future research. Future analysis might
also focus on how different rating outcomes for the very same CDO tranche can be
linked to different CDO rating processes and applied models.
92
Impact of Multiple CDO Ratings on Credit Spreads
Reflecting the results of our analysis with regard to the current discussion on required
regulation of markets and rating agencies in particular, we interpret them as a support
for the argument that the crisis was caused by misaligned incentives and ensuing intransparency. Accordingly, investors should not only request for higher credit spreads
in opaque situations but also demand transparency and thereby induce the development of sophisticated incentive structures.
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
93
6 Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
6.1 Introduction
Throughout the process of asset allocation investors often rely on information provided by third parties in order to overcome existing information asymmetries. In this
context it is widely acknowledged that stock investors rely on stock analysts’ forecasts
whereas investors in credit-linked instruments often delegate monitoring activities to
rating agencies. Each interaction follows a principal-agent relationship with the stock
analyst and the rating agency respectively acting as agents on behalf of the investor
(principal). Since stock analysts’ forecasts are publicly available information, we argue
in the following that the impact of (changing) stock analysts’ forecasts should not be
limited to stock prices, but should also be taken into consideration by investors of
credit-linked instruments. In turn, credit-linked instruments incorporate valuable information on the underlying reference entities and are also publicly accessible to stock
analysts throughout the process of reviewing their own estimates. From an economic
perspective, it is therefore worthwhile to investigate dynamic structures and spill-over
effects between these two adjacent capital markets or market participants respectively.
In the following we focus on credit default swaps (CDS) and stock analysts’ earnings
forecasts and investigate the relationship between the two in relation to short-term as
well as long-term dynamic structures. Thus, the chapter’s scope is not limited to comovement but also takes into account lead-lag structures. CDS markets were selected
since they offer both comparable high levels of liquidity and are viewed to be an important benchmark to assess credit risk.
Based on a data sample of 204 reference entities from Europe and the U.S., we analyze
interacting structures between CDS spreads and mean stock analysts’ forecasts of
earnings-per-share (FEPS) on a monthly basis covering the period December 2003 until December 2008. Typically, not one but several stock analysts estimate the future
earnings of a particular company. Thus, it is possible to calculate the corresponding
mean values. We chose earnings-per-share (EPS) estimates since they are the most
frequently forecasted earnings benchmark by stock analysts and thus offer the largest
94
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
data pool with regard to our research efforts. Application of EPS estimates is also a
widely accepted standard within financial literature on stock analysts’ forecasts. For
the purpose of our empirical analysis we obtained CDS spreads from Bloomberg and
stock analysts’ forecasts from Institutional Brokers Estimate System (I/B/E/S). Both
data sources have been largely used by prior authors of studies on CDS spreads and
stock analysts’ forecasts respectively (e.g. Diether et al., 2002). Against the background of a correlation analysis we illustrate that higher CDS spreads (monthly average changes) are associated with lower stock analysts’ earnings forecasts (monthly average changes). By applying a panel regression as well as a vector auto regression
analysis (VAR) we show that neither CDS markets are leading stock analysts’ forecasts nor the latter the first.
In addition to the dynamics that exist between CDS spreads and stock analysts’ earnings forecasts we also analyze the co-movement and lead-lag structures between CDS
spreads and the dispersion underlying mean stock analysts’ earnings forecasts. Dispersion of stock analysts’ forecasts is defined as the ratio of standard deviation of all outstanding stock analysts’ forecasts and the absolute value of the corresponding mean
forecast. Our findings show that CDS spreads and dispersion of stock analysts’ forecasts are positively correlated with CDS spreads actually leading the formation of dispersion of stock analysts’ forecasts. This also documents the results of a cointegration
test allowing the detection of long-run equilibrium relationships between dispersion of
stock analysts’ forecasts and CDS spreads. Additionally, we find significantly different
patterns between U.S. and European entities. On average, the spill-over effects between CDS spreads and dispersion of stock analysts’ forecasts seems to be more pronounced in the case of U.S. reference entities than for European CDS contracts.
So far, the existing financial literature on the co-movement and the lead-lag structures
of the CDS markets has been limited to its interdependence with either rating changes
or the stock and bonds markets (see for example Blanco et al., 2005; Daniels and Jensen, 2005; Lehnert and Neske, 2006; Norden and Weber, 2007). Similarly, research
targeting the stock analysts’ forecasts is primarily limited to its exchange with the
stock markets (see for example Diether et al. 2002; Johnson, 2004; Doukas et al.,
2006). The ensuing empirical analysis therefore places us in a niche between the broad
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
95
research strings on CDS markets and stock analysts’ forecast, and aims to provide a
missing link in the literature between the CDS markets and stock analysts’ as important information agents on the worldwide capital markets.
The chapter is organized as follows: Based on a literature review in chapter 6.2, we
present the basic concept of spill-over effects between dispersion of stock analysts’
earnings forecasts and develop three different hypotheses with regard to co-movement
and lead-lag structures (chapter 6.3). We provide a comprehensive overview of the applied data (chapter 6.4) and throughout chapter 6.5 present different test statistics and
robustness checks in order to verify the hypotheses initially formulated. Chapter 6.6
concludes the research on spill-over effects.
6.2 Literature Review
At the heart of the analysis is the co-movement between the CDS markets and stock
analysts’ forecasts as well as their respective responsiveness. Against this background,
the following literature review is grouped around: (1) the existing research on the comovement of the CDS markets with adjacent asset classes and information agents; and
(2) the impact of the (dispersion of) analysts’ earnings forecasts on the bond and stock
markets.
Referring to (1), the idea of the credit markets’ co-movement with endogenous information agents initially goes back to Katz (1974) and Weinstein (1977), who are among
the first to compute the responses of the bond markets based on the activity of information agents (credit rating changes). Katz (1974) shows that rating adjustments are
not anticipated by the market participants, and that the bond markets need up to ten
weeks before the rating adjustments are fully reflected in the underlying bond prices.
Weinstein (1977), in turn, disagrees with Katz (1974) by showing that rating adjustments are anticipated by the markets for a period from 7 to 18 months prior to the rating changes, but no or only a few reactions are anticipated during or after the rating
adjustments. Holthausen and Leftwich (1986) further differentiate the rating adjustments into up- and downgrades, and investigate their impacts on the corresponding
stock prices. They outline that downgrades are linked with negative abnormal stock
returns (in a two-day window) and that no significant positive abnormal stock returns
96
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
occur in the case of an upgrade. Both for the bond and stock markets, other studies also find empirical evidence for rating downgrades leading to stronger responses (negative returns) as observed in the case of rating upgrades (e.g. Hand et al., 1992; Goh and
Ederington, 1993; Kliger and Sarig, 2000). Often, no significant abnormal returns are
detected for rating upgrades (e.g. Dichev and Piotroski, 2001).
Norden and Weber (2004) analyze with regard to the CDS markets whether prices
react after a rating event, based on the assumption that credit ratings convey new information to the market. If credit ratings only reflect information that is already known
by the market, the prices should not react to the rating event at all. They conclude, inter alia, that both the CDS and the stock market anticipate rating downgrades. Hull et
al. (2004), in turn, focus on the interactions between the CDS and credit rating announcements and observe that reviews for downgrades contain significant information,
whereas, in contrast to Norden and Weber (2004), the actual downgrades do not. Hu
and Cantor (2006) investigate the dependency between the credit issuance spread and
level of rating and thereby document a positive correlation between credit spreads and
decreasing credit quality for a data set of 16,516 structured finance transactions. In
their paper on the effect of credit ratings on CDS spreads and credit spreads Daniels
and Jensen (2005) conclude that CDS spreads and credit spreads are indeed correlated
but unequal, on average. They document changes in CDS spreads (negative returns) in
the case of downgrades, but no statistically significant reactions are observed in the
case of upgrades. In addition, they found empirical evidence of lag structure, which
can be aligned to earlier research relating to the interchange between the bond markets
and rating announcements (e.g. Weinstein, 1977). Finally, Daniels and Jensen (2005)
indicate that rating adjustments are anticipated by the CDS markets. This evidence is
not supported by Lehnert and Neske (2006) for European reference entities. All of the
studies documented negative returns in the case of rating downgrades and were discordant on the question of whether the stock and CDS markets anticipate the rating announcements or not. In turn, the direct correlation between the CDS markets and (traditional) bonds markets, without specifically addressing the issue of credit ratings, has
also been covered by academics. Blanco et al. (2005) demonstrate, for example, that
the CDS market leads the bond market in terms of defining the market price for credit
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
97
risk (see also Longstaff et al., 2003). Norden and Weber (2007) provide us with a deep
analysis of the co-movement between CDS and stock markets. Their results indicate
that lagged stock returns impact CDS spreads and they argue that stock returns thus
lead CDS spreads. In addition, Benkert (2004) uses earnings-based accounting figures
as impact factors in order to explain the default swap premium. He succeeds in documenting the significant impact of these figures on CDS spreads.
With regard to rating agencies and stock analysts, Ederington and Goh (1998) focus on
the “who follows whom” patterns. Testing for the bond and stock market, they observe
that downgrades are preceded by lower earnings and forecasts expectations on the analysts’ side, and that future forecasts even tend to continue falling in the course of
downgrades. Rating upgrades, in turn, are followed by higher analysts’ forecasts.
Research on stock analysts’ forecasts is the second string of financial literature this
chapter relies on and can be structured according to studies on forecast accuracy (e.g.
Fried and Givoly, 1982; Patz, 1989; Bolliger, 1998; Brown, 2001; Capstaff et al. 2001;
Richardson et al., 2004; Cowen et al., 2006; Oster, 2007; Balboa et al., 2008), the effect on stock prices, as well as the dispersion of analysts’ forecasts. The latter two will
be discussed in more detail, as they are crucial for deriving the competing hypotheses
throughout the subsequent chapter. As already outlined, one of the most common
measures in this context are earnings forecasts, generally displayed as EPS.
As one of the first, Abdel-Khalik and Ajinkya (1982) succeed in detecting positive abnormal stock returns in the course of their revisions leading to higher analysts’ forecasts and negative abnormal returns for revisions towards lower analysts’ forecasts.
Their finding are confirmed by several additional studies on the impact of analysts’
forecasts on stock prices (e.g. Peterson and Peterson, 1982; Lys and Sohn, 1990;
Stickel, 1991; Asquith et al., 2005). In the context of the dispersion of analysts’ forecasts, stock analysts in turn are viewed as a proxy for the behavior of investors and
thus dispersion provides insights into the diversity of investors’ beliefs (Morse et al.,
1991), whereas dispersion represents the idea that stock analysts or investors, respectively, do not agree about future expectations. Thus, the dispersion of the stock analysts’ forecasts is often assumed to be a measure of uncertainty (Miller, 1977). A variety of studies analyzed the relationship between (future) stock returns and the disper-
98
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
sion of analysts’ forecasts, and documented the negative relationship between dispersion and future returns. Specifically, the empirical design uses the current dispersion of
analysts’ forecasts of future earnings and applies it to future stock returns. Thus, the
high dispersion in analysts’ forecasts leads to lower future returns or vice versa (e.g.
Ajinka et al., 1991; L’Her and Suret, 1996; Ackert and Athanassakos, 1997; Dische,
2002; Baik and Park, 2003; Park, 2005). In contrast to these empirical findings, few
other studies present a positive relationship between dispersion and future returns but
instead produce rather opposing results (e.g. Kazemi, 1991; Doukas et al. 2006). Based
on these different empirical outcomes, a discussion evolved among academics regarding whether the dispersion of the analysts’ forecasts is a measure for risk (in the case
of a positive correlation with future stock returns) or a measure of the differences in
investors’ opinions (in the case of a negative correlation with future stock returns).
Against the background of this dispute, Diether et al. (2002) perform one of the most
comprehensive analyses on the dispersion of analysts’ forecasts and find additional
support for the hypothesis that dispersion can be regarded as a proxy for differences of
opinion. On the basis of a data sample for US stocks (1983-2000), Diether et al. (2002)
confirm that stocks with a high dispersion of analysts’ forecasts earn significantly lower future returns than are observed otherwise. However, following the empirical outcomes of Diether et al. (2002), Johnson (2004) explains the negative relationship between dispersion and future stock returns with unpriced information risk. The dispersion of the analysts’ forecasts in this framework acts as a proxy for unpriced information risk.
Thus, the existing financial literature on the co-movement of the CDS markets is limited to its interdependence with either rating changes or the stock and bonds markets.
Research targeting the stock analysts’ forecasts is, in turn, primarily bounded to its exchange with the stock markets. The following empirical analysis therefore places us in
a niche between the research on CDS markets and stock analysts’ earnings forecasts
respectively, and thus aims to provide a missing link in the financial literature between
the CDS markets and stock analysts acting as information agents on the worldwide
capital markets.
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
99
6.3 Spill-over Effects between Stock Analysts’ Forecasts and CDS
Spreads
6.3.1 Mean Stock Analysts’ Forecasts and CDS Spreads
Stock analysts are seen as information agents for investors in relation to stock investments. Throughout the investment process, stock analysts’ reports as well as stock analysts’ forecasts offer valuable guidance to the investor. A positive correlation between
forecasts and future stock returns empirically confirms the incremental information
provided by the stock analysts to investors through their forecasts (e.g. Peterson and
Peterson, 1982; Lys and Sohn, 1990; Stickel, 1991; Asquith et al., 2005). In the case of
rating agencies and creditors, the relationship is primarily about delegated monitoring
activities. However, stock investors eventually delegate due diligence activities to the
corresponding stock analysts, too. Using a principal-agent framework in order to describe the interaction between stock analyst and investor, we note that the investor acts
as a principal delegating due diligence activities towards the agent (stock analysts).
Even so, that does not necessarily lead to direct compensation for stock analysts. Specifically, investment banks offer their clients free access to analysts’ reports in exchange for brokerage business. However, there is at least an indirect compensation
scheme in place.
In the context of a CDS contract, the principal invests and agrees to bear a certain credit risk in exchange for an annual payment, whereas the investor acts as a so-called
protection seller and the issuer as a protection buyer. The amount of annual payment is
calculated on the basis of CDS spreads, which in turn are displayed in bps. With regard to the level of CDS spread, we generally state that higher credit risk is positively
correlated with higher CDS spread levels. Common examples of a protection buyer are
banks hedging their own portfolio’s credit risk. In contrast to the stock markets, the
CDS markets, by definition, rather focus on the underlying credit risk. This perception
is supported by existing research on the CDS markets, which subsequently puts a
strong focus on the rating agencies and rating changes (e.g. Norden and Weber, 2004).
In this context, rating agencies perform monitoring activities on behalf of the investor
and can thus be viewed as agents, with the investor as a corresponding principal. Since
100
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
it is the issuer who pays the rating agencies, there is again no direct compensation
scheme in place. However, these costs are ultimately borne by the investor and an indirect compensation can be detected.
In order to explain the spill-over effects between the CDS markets and stock analysts’
forecasts we merge these two adjacent markets in the following. CDS spreads are a
measure of the riskiness of the underlying corporate debt of a specific reference entity,
and stock analysts’ forecasts, in turn, can be viewed as a proxy for the market expectation of the future prospect of the very same reference entity. Against this background,
the applied theoretical model rests on a principal-agent framework combining the
perspective of both the CDS market as well as the stock market, with the investor as an
interacting element. In detail, we assume that the investor is active in both submarkets
and does not ignore the information provided through stock analysts when making investment decisions with an exposure to the CDS market. Even if we assume that the
investor is actively trading in the CDS segment only, this does not mean that he is cut
off from access to the analysts’ forecasts, as they are publicly available. Rather opposing to the idea of correlation structures between CDS markets and stock analysts’ estimates are different underlying dimensions: CDS spreads actually account for the
riskiness of underlying credit risk, whereas EPS estimates account for the profitability
of equity investments. However, even if EPS estimates address a performance-driven
dimension and CDS spreads are rather focused on a default-driven dimension, we argue in the following against the background of an asset-based credit risk model as introduced by Merton (1974) that changes in company’s asset levels also affect its default probability. Merton’s model derives a company’s default risk foremost from the
volatility of the entity’s value, which in turn is of course affected by earnings- or performance-driven dynamics respectively. Hence, any changes in stock analysts’ forecasts should impact the CDS spread levels and vice versa. We thus argue that:
Hypothesis H1:
Mean stock analysts’ earnings forecasts should be negatively correlated with CDS spreads.
With the co-movement between the stock analysts’ earnings forecasts and CDS markets at the very heart of this chapter, we also focus on lead-lag structures or “who-
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
101
follows-whom” patterns between earnings forecasts and CDS spreads. If investors are
indeed influenced by changes in the stock analysts’ earnings forecasts - against the
background of their exposure to the CDS market - we view the changes in the analysts’ forecasts as an additional independent pricing factor of future CDS premiums
(e.g. Benkert, 2004). In addition to the spill-over effects induced by the stock analysts’
forecasts, it is also conceivable that the CDS markets actually influence the stock analysts and their forecasts subsequently, as observed for CDS markets in relation to rating changes. The idea of CDS markets actually leading stock analysts’ forecasts can
also be aligned to existing literature on CDS spreads and rating changes documenting
that CDS markets actually anticipate rating changes (e.g. Norden and Weber, 2005).
This argument primarily relies on different time settings in which CDS spreads and
analysts’ forecasts are updated or reviewed, whereas CDS spreads are traded on a daily
basis. This kind of adjusting mechanism does not hold for mean stock analysts’ forecasts. Mean earnings forecasts are in fact monthly publicized, but represent an aggregated perspective summarizing the forecasts of a number of different analysts made
throughout a given month. In contrast to the case of anticipating rating changes, it is
rather difficult to assign an event study approach allowing to control for lead-lag structures on a daily level. As we use monthly observations rather than daily data points
throughout the subsequent analysis, we believe that new information regarding a reference entity impacting earnings forecasts or credit risk levels are reflected in the corresponding monthly time series without any lags. This argument is also aligned to Fama’s (1991) statement that capital markets are efficient and market participants continuously adjust to new information. Thus, we subsequently test for the hypothesis that
lead-lag relationships between CDS premiums and stock analysts’ forecasts exist, and
assume that neither CDS spreads nor stock analysts’ forecasts lead on a monthly basis
the other. Accordingly, we propose that:
Hypothesis H2:
There are no lag structures between mean stock analysts’ earnings
forecasts and CDS spreads.
102
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
6.3.2 Dispersion of Mean Stock Analysts’ Forecasts and CDS Spreads
Besides the analysis of the direct lead-lag structures between the CDS markets and
analysts’ forecasts, this chapter also aims to analyze to what degree uncertainty among
analysts is reflected in the pricing structure of CDS contracts. We would like to point
out that it is not a necessity that the co-movement of CDS spread levels with either the
stock analysts’ forecasts or the dispersion of the stock analysts’ forecasts follows exactly the same patterns. Even if the stock analysts’ forecasts (e.g. mean EPS forecasts)
vary over time, the underlying dispersion may remain stable or vice versa.
Again, stock analysts are viewed as a proxy for the behavior of investors and thus dispersion gives insights into the diversity of investors’ beliefs (Morse et al., 1991), whereas dispersion represents the idea that stock analysts or investors disagree about future
expectations. Thus, the dispersion of stock analysts’ forecasts is often assumed to be a
measure of uncertainty (Miller, 1977). However, the direct application of Miller’s
(1977) stock market-based argumentation to a CDS environment is rather difficult, as
it centers on the non-existence of short selling, which in turn has no equivalent on the
CDS markets. Nevertheless, we view the dispersion of analysts’ forecasts as a proxy
for uncertainty and assume that stock analysts act as information agents on behalf of
investors. Higher dispersion of stock analysts’ forecasts can be viewed as an indication
of increased earnings volatility making future earnings of a reference entity less predictable. Thus, if future earnings are less predictable – again based on Merton (1974) –
the default probability of the underlying reference entity should increase and ultimately should lead to higher CDS spread levels subsequently. Faced with less predictable
future earnings, investors, who are willing to act as protection sellers, will demand a
higher credit spread in order to compensate for the incurred credit risk underlying the
CDS contract as observed in the case of predictable future earnings of otherwise similar reference entities. In addition, the gradient disagreement between their information
agents may also lead to an increased willingness subsequently to hedge against the future credit risk of the corresponding reference entities. Such behavior in turn triggers
demand on the side of the CDS protection buyers and induces a higher future CDS
premium. Based on these considerations, it is hypothesized that:
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
Hypothesis H3:
103
Dispersion of mean stock analysts’ earnings forecasts should be
positively correlated with CDS spreads.
Furthermore, we analyze the level of co-movement between the dispersion of stock
analysts’ forecasts and CDS spreads with regard to lead-lag relationships. Throughout
the argumentation leading to Hypothesis 3 we argue from an investor’s angle. In this
context lead-lag structures do not play any meaningful role at all, because investors
notice increasing dispersion of stock analysts’ forecasts and adjust their risk/return
profile with regard to risk premiums on CDS markets immediately. Throughout the
following argumentation we now focus on the perspective of stock analysts. As outlined earlier on, CDS contracts are priced on a daily basis, whereas stock analysts review their forecasts less frequently. This allows stock analysts to observe market
movements with regard to CDS spreads before they actually compile their forecasts. In
line with Hypothesis 2 we do not assume that analysts receive any new information out
of CDS spreads with regard to the underlying reference entity which they do not have
incorporated into their forecasts anyway at the same time. However, we believe that
CDS spreads incorporate valuable information with regard to future earnings volatility
as assumed by market participants impacting disagreement between stock analysts:
Higher CDS spreads correspond to higher levels of credit risk, which in turn - according to Merton (1974) - can be assessed as higher asset volatility. Higher asset volatility
finally displays higher earnings volatility or less predictable future earnings streams
leading to increasing disagreement between stock analysts. Again, the definition of
changes in CDS spreads is in this context extended beyond the mere reflection of being a function of new (positive/ negative) company news but to a proxy for earnings
volatility. Since stock analysts’ forecasts typically reflect upon a on longer time period
we believe that they also apply a long-term perspective on CDS times series rather
than a focus on daily changes in order to assess regime switches with regard to earnings volatility accordingly. This leads us along the above argumentation to:
Hypothesis H4:
CDS spreads lead the dispersion of stock analysts’ earnings forecasts.
104
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
6.4 Data Sample
The empirical part of this chapter relies on CDS spreads and consensus EPS forecasts
for 204 different reference entities. It covers the period from December 2003 until December 2008 or 61 monthly observations respectively (see Appendix I). CDS spreads
correspond to the observed individual 5 year CDS spread levels and were derived from
Bloomberg. The distinction of “5 years” represents the maturity of the analyzed CDS
contract. we chose the 5-year CDS contract as it attracts the highest liquidity, in comparison to other maturity structures, and thus reflects the most current quote conditions. Each reference entity either belongs to the CDX Investment Grade Series, CDX
High Yield Series, iTraxx Europe Series or iTraxx Crossover Series. Affiliation to one
of the four main worldwide CDS indices is viewed as a prerequisite to be included in
the subsequent analysis in order to guarantee a comparably high level of liquidity between the reference entities. This selection condition should allow us to reduce the likelihood of potential size and/or volume bias in the empirical results: the analyzed reference entities belong to the worldwide largest companies and are all viewed to be socalled blue chips with their stocks being included in one of the major stock market indices. In addition to CDX and iTraxx, we also differentiate between Investment Grade
(CDX Investment Grade Series, iTraxx Europe Series) and High Yield (CDX High
Yield Series, iTraxx Crossover Series). We started the data analysis by looking at the
current and past composition of the above mentioned four CDS sub-indices and compiled the corresponding CDS spreads, as displayed by Bloomberg. The rationale to include past index affiliates is to minimize the threat of ex-post selection bias. As we
will show in the following, due to the missing data points in the time series, we are unable to reflect fully the issue of selection bias with regard to bankruptcy. However, this
procedure leads us to an initial figure of 504 different reference entities. In a second
step, we exclude all reference entities for which we did not obtain a complete time series for the period from December 2003 until December 2008. We intentionally exclude CDS reference entities from the sample with missing data in order to guarantee
stable results throughout the subsequent time series analysis. The period from December 2003 until December 2008 was chosen in order to achieve both a statistically reliable times series and robust sample size in terms of included reference entities. In addi-
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
105
tion, we allocate the corresponding stock analysts’ forecasts to the identified CDS time
series, which in turn lead us to the definitive data sample of 204 different reference
entities. Allowing for missing data (both stock analysts’ forecasts as well as CDS
spread levels), we finally came up with a data sample of 204 different reference entities for which we were able to allocate both the CDS spread levels as well as the stock
analysts’ forecasts throughout the period from December 2003 until December 2008.
This leaves us with 61 monthly observations each. One reason for such a high number
of excluded reference entities lies in the fact that the data was merged together from
three different information sources (CDS index affiliation, Bloomberg and I/B/E/S).
Take, for example, the case of a company which is taken to be private throughout the
analyzed period. In this case, the CDS contracts are still traded and spread levels are
displayed, but the stock analysts stop covering this specific company and we have to
exclude it from the data sample. However, with 204 different reference entities, the
applied data sample of our empirical analysis proves to be considerably higher in
comparison to the existing empirical literature on CDS markets (e.g. Norden and Weber (2007) only use a total of 58 reference entities throughout their analysis).
CDS spreads are indicated in basis points (bps). Stock analysts’ earnings forecasts
were derived from the I/B/E/S data base, which is the standard data source for empirical studies on stock analysts’ forecasts (see, for example, Diether et al., 2002). As a
measure of the stock analysts’ forecasts, we chose average EPS forecasts compiled by
I/B/E/S for each reference entity on a monthly basis and define them in the following
as FEPS. We selected FEPS, as it is the most frequent accounting figure forecasted by
stock analysts and thus proves to be the most complete data set. FEPS are displayed in
the local currency of the corresponding entity, which is either EUR or USD in more
than 95% of cases in the data sample. I/B/E/S allocates the forecasts of each analyst on
a monthly basis for each specific company. On this basis, the I/B/E/S calculates and
displays the mean (consensus) FEPS for each company. Based on the publication data,
we merged the derived mean FEPS with the corresponding CDS spread on the specific
day (last quote at which a trade was settled). As the I/B/E/S typically releases new
forecasts around the 20th of each month, we are able to comply with a monthly routine
between two data points. Of course, FEPS are always calculated with respect to a spe-
106
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
cific forecast period. For the benefit of the empirical analysis, we chose as the benchmark period (t+1) the period following the most recent published actual earnings per
share (t). Given, for example, a fiscal year ending in December and the latest actual
EPS reported by a specific company in December 2007 (t), the benchmark period of
the FEPS used throughout analysis would be the estimated earnings for the fiscal year
ending in December 2008 (t+1). The preceding fiscal year ending is the most precise
description of the current economic situation of a company in terms of the stock analysts’ forecasts and thus is the most applicable counterpart to CDS spreads, which in
turn measure the company’s current credit risk.
In Table 6-1 we present an overview of the data sample applied throughout the empirical part of this chapter. Of the 204 different reference entities, 113 (55.4%) are from
the CDX index family and 91 (44.6%) from the iTraxx index family. In addition, the
majority of entities (87.7%) belong to the sub-index “Investment Grade”. Additionally,
we note that the sum of the two sub-indices, “High Yield” (28) and “Investment
Grade” (180) does not equal the corresponding total figures (204). This gap is a simple
result of the fact that some reference entities have changed affiliation between the two
sub-indices throughout the analyzed time period (e.g. General Motors started in the
CDX Investment Grade Series but eventually ended up in the CDX High Yield Series).
In order not to change the applied search algorithm, we decided to include entities that
changed index affiliation in both sub-indices “Investment Grade” and “High Yield”.
Of course, on a total scale we only counted them once. Besides the mean values of
CDS spreads and stock analysts’ forecasts, we also display the mean values of the corresponding standard deviation (SD) of the stock analysts’ forecasts. Typically, not one
but several stock analysts estimate the future earnings of a particular company. Thus,
the standard deviation is computed on a monthly basis against the background of different stock analysts’ forecasts. In order to generate comparable levels of standard
deviation, we defined SD as a ratio of observed mean standard deviation and absolute
mean stock analysts’ EPS forecasts. Throughout this chapter, we define the dispersion
of stock analysts’ forecast as the above-mentioned SD and use both terms synonymously. Besides the mean values of CDS Spread, FEPS and SD, we also display the
monthly average changes of CDS spreads, FEPS and SD and label them as ¨&'6
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
107
Table 6-1: Overview of Mean Analysts’ Forecasts and CDS Spreads
Mean Values
2004*
2005
CDS Spread
214.88
294.49
FEPS
2.25
1.52
SD**
0.25
0.22
ѐCDS Spread
-0.01
0.10
ѐFEPS
-0.08
-0.02
ѐSD
0.30
0.39
# of entities
12
12
Investment Grade
CDS Spread
52.61
58.56
FEPS
2.49
2.93
SD**
0.08
0.05
ѐCDS Spread
-0.01
0.05
ѐFEPS
0.02
0.02
ѐSD
0.26
0.19
# of entities
104
104
Total (CDX)
CDS Spread
66.98
70.84
FEPS
2.40
2.83
SD**
0.10
0.06
ѐCDS Spread
-0.01
0.05
ѐFEPS
0.01
0.02
ѐSD
0.27
0.19
# of entities
113
113
iTraxx
High Yield
CDS Spread
212.46
161.90
FEPS
2.60
2.89
SD**
0.72
0.67
ѐCDS Spread
0.00
0.03
ѐFEPS
0.54
0.36
ѐSD
0.34
0.24
# of entities
16
16
Investment Grade
CDS Spread
36.41
30.78
FEPS
2.26
2.52
SD**
0.26
0.35
ѐCDS Spread
-0.01
0.02
ѐFEPS
1.05
0.04
ѐSD
0.08
0.22
# of entities
76
76
Total (iTraxx)
CDS Spread
67.19
53.68
FEPS
2.34
2.61
SD**
0.35
0.41
ѐCDS Spread
-0.01
0.02
ѐFEPS
0.97
0.10
ѐSD
0.13
0.22
# of entities
91
91
Total
High Yield
CDS Spread
213.50
218.73
FEPS
2.45
2.31
SD**
0.52
0.48
ѐCDS Spread
-0.01
0.06
ѐFEPS
0.27
0.20
ѐSD
0.32
0.30
# of entities
28
28
Investment Grade
CDS Spread
45.77
46.83
FEPS
2.39
2.76
SD**
0.16
0.18
ѐCDS Spread
-0.01
0.04
ѐFEPS
0.45
0.03
ѐSD
0.19
0.20
# of entities
180
180
Total
CDS Spread
67.07
63.18
FEPS
2.37
2.73
SD**
0.21
0.22
ѐCDS Spread
-0.01
0.04
ѐFEPS
0.44
0.05
ѐSD
0.21
0.21
# of entities
204
204
* Values for 2004 also include December 2003
** Ratio of computed standard deviation and absolute mean forecasts of EPS
CDX
High Yield
2006
338.21
1.47
0.51
-0.03
-0.17
0.60
12
56.03
3.41
0.09
-0.03
0.00
0.21
104
68.14
3.27
0.09
-0.03
0.00
0.20
113
125.56
2.73
0.44
-0.03
0.09
0.13
16
28.17
2.85
0.18
-0.02
0.13
0.15
76
45.10
2.85
0.22
-0.02
0.12
0.14
91
216.70
2.19
0.47
-0.03
-0.02
2.55
28
44.26
3.17
0.12
-0.03
0.05
0.18
180
57.86
3.08
0.15
-0.03
0.05
0.18
204
2007
255.09
1.89
0.18
0.04
-0.06
0.72
12
60.80
3.46
0.07
0.11
0.00
0.29
104
69.75
3.34
0.07
0.10
0.00
0.28
113
107.63
3.06
0.40
0.08
0.25
0.14
16
30.76
3.23
0.17
0.16
0.23
0.07
76
43.89
3.22
0.21
0.15
0.24
0.08
91
170.83
2.52
0.31
0.06
0.12
0.39
28
48.12
3.36
0.11
0.13
0.10
0.19
180
58.22
3.29
0.13
0.12
0.11
0.19
204
2008
882.64
2.25
0.40
0.19
0.07
0.86
12
227.92
3.63
0.14
0.17
-0.01
0.34
104
251.17
3.51
0.14
0.17
-0.01
0.39
113
389.48
2.91
0.70
0.17
0.27
1.17
16
125.22
3.27
0.24
0.18
0.19
0.25
76
169.55
3.23
0.32
0.18
0.21
0.41
91
600.83
2.63
0.58
0.18
0.19
1.03
28
184.56
3.48
0.18
0.17
0.08
0.30
180
214.76
3.39
0.22
0.17
0.09
0.40
204
Total
394.08
1.86
0.31
0.06
-0.05
0.57
12
90.55
3.17
0.08
0.06
0.01
0.26
104
104.75
3.06
0.09
0.06
0.00
0.27
113
199.62
2.83
0.59
0.05
0.31
0.40
16
50.04
2.82
0.24
0.06
0.34
0.15
76
75.74
2.84
0.30
0.06
0.34
0.20
91
282.96
2.42
0.47
0.05
0.15
0.91
28
73.45
3.02
0.15
0.06
0.15
0.21
180
91.81
2.96
0.19
0.06
0.15
0.24
204
108
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
¨)(36 DQG ¨6' 6FDOHG E\ WKH WLPH GLPHQVLRQ - 2008), we observe CDS
spreads to rest, on average, throughout the years 2004 - 2007 in a comparable spread
interval, reaching a floor in 2006/7. However, the financial turmoil of 2008 led to significantly higher CDS spread levels on average. As expected, spread levels for the subindex “High Yield” are higher, as observed in the case of “Investment Grade”. A closer inspection also reveals that on average “iTraxx” affiliated entities achieve lower
spread levels as observed in the case of “CDX”. However, a reliable comparison
would only be achieved by crosschecking the individual credit quality, for example,
against the corresponding ratings of the different index entities. Otherwise, the spread
difference can be explained as a mere result of different credit quality on average in
the two index families.
Significantly higher levels of dispersion of the analysts’ forecasts in terms of European
reference entities raise some questions with regard to the data quality of I/B/E/S. As
I/B/E/S was introduced to the US market first, we cross-checked the dispersion of the
analysts’ forecasts as displayed by I/B/E/S with a second database (compiled by the
JCF Group), which primarily focuses on the European stock markets. The results confirm the significantly higher levels of dispersion of the analysts’ forecasts as generated
on the basis of I/B/E/S in the first place. Because mean FEPS are biased both by the
number of outstanding shares as well as by the currency accounted for the specific reference entity, simply scaled by the years, they do not provide us with additional crucial
information. Thus, a far more revealing aspect are the relative values of FEPS (average
monthly changes). Despite the fact that on average CDS spread levels notably increased throughout 2008 (17%), FEPS are also still increasing. This leaves us with a
first sign indicating the possibility of existing lag structures. In fact, throughout the
whole data sample, the stock analysts prove to be very optimistic, predicting increasing future earnings in nearly all of the observed cases. Nevertheless, we have to take
into account the fact that the analyzed period coincides with a period of overall economic growth (e.g. high worldwide GDP growth rates). SD seems to be positively correlated with CDS spread levels. Thus, a higher level of perceived credit riskiness is associated with a high degree of disagreement between stock analysts.
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
109
6.5 Empirical Results
6.5.1 Co-movement
Having detected first signs of co-movement between the stock analysts’ forecasts and
CDS spreads on the basis of a mere comparison of the mean values, we now perform a
series of correlations using Pearson correlation coefficients in order to address the issue of co-movement in more detail. We retain the sorting of the data sample, as outlined in Table 6-1, but for reasons of clarity the ensuing correlation analysis is limited
to the two index families, CDX and iTraxx. For each of the two index groups as well
as for the total sample, we illustrate in Table 6-2 the mean values of correlation coeffiFLHQWVȡEHWZHHQWKHYDULDEOHV&'6t-l, FEPSt-l as well as SDt-l, with l indicating the corresponding lag structures (l = 1 … 3). In the case of no lag structures (l = 0) we simultaneously refer to CDSt, FEPSt as well as SDt. Computed mean values of correlation
FRHIILFLHQWVȡLQWXUQUHO\RQILUP-specific analysis of correlation on the level of each
reference entity i. In order to capture both dimensions of lead-lag structures between
two variables, two correlation series have to be calculated. Based on these considerations, we finally come up with 144 different correlation coefficients. In addition, we
apply a t-WHVWVWDWLVWLFUHO\LQJRQDVLJQLILFDQFHOHYHORIĮ LQRUGHUWRGHWHUPLQH
ZKHWKHUWKHREVHUYHGȡ’s are significantly different from zero or not. Since – as outlined before – the absolute values of FEPS might be affected by size as well as currency biases, we focus in the following primarily on the correlation coefficients relying on
monthly changes of CDSt-l (¨&'6t-l), FEPSt-l (¨)(36t-l) and SDt-l (¨6't-l) as observed
for all reference entities on average. For reasons of completeness, we DOVRUHSRUWȡ’s
for absolute values.
With regard to the correlation structures between CDSt-l and FEPSt-l, we observe different correlation patterns for CDX-linked in the case of absolute values, as documented for iTraxx-linked reference entities. On an un-lagged level we find no particular correlation patterns in the case of CDX entities HJȡ&'6t, FEPSt) = 0.0016). For
iTraxx-linked entities, we actually observe results opposing the argument leading to
Hypothesis 1 HJȡ&'6t, FEPSt) = 0.1553), which are also confirmed by CDX entities for lagged correlation coefficients (HJ ȡ&'6t, FEPSt-3) = 0.0837). However,
110
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
¨&'6t-l and ¨)(36 t-l follow comparable negative patterns both for CDX and iTraxx
index families and therefore support Hypothesis 1. Lead-lag structures, as observed in
Table 6-2, for the average ¨&'6 t-l and ¨)(36 t-l seem to be more revealing in the case
of CDS spreads leading FEPSs (ȡ)(36t, CDSt-l)). Again, Table 6-2 illustrates differences with regard to the index affiliation of the reference entities. Whereas CDXaffiliated assets put a slightly stronger emphasize on the idea of CDS spreads leading
future stock analysts’ forecasts; the iTraxx-affiliated assets indicate higher levels of
significance for stock analysts’ forecasts actually leading the CDS spreads. However,
we must consider the rather asymmetric increase in FEPS in the case of increasing
CDS spread levels, as documented for the subgroup “iTraxx”. In most cases, the significance of the correlation coefficients increases with the widening lag structures and
maximizes with a lag of l = 2 or l = 3 respectively. Nevertheless, we need to take into
account the fact that the scope of the correlation coefficients is rather limited to assessing inter-temporal co-movements between time series.
A second string of correlation structures evolves between CDS spreads and the corresponding SD. On a total sample level, DVZHOODVIRUERWKLQGH[JURXSȡ¨&'6 t-l, ¨6' tl),
this proves to be significantly positive. A positive correlation is also confirmed on
the level of absolute values for CDSt-l and SDt-l. Direct comparison of ȡ&'6t-l, SDt-l)
and ȡ¨&'6t-l, ¨6' t-l) reveals that levels of significance are on average higher given
the analysis of absolute values. This is particularly true in the case of CDS spreads
leading SD (ȡ&'6t-l, SDt)). However, in the case of ȡ¨&'6t-l, ¨6' t-l) we detect significant non-lagged co-movement for ȡ¨&'6t, ¨6' tDVZHOODVIRUȡ¨&'6 t-1, ¨6' t)
with the later supposing that CDS spreads actually lead the composition of SD. In addition IRU ȡ&'6t-l, FEPSt-l) we observe that the positive correlation is significantly
higher in case of CDX-linked credit spreads as in turn displayed for iTraxx-linked
entities. With ¨&'6 t-1 actually leading ¨6' t and ¨)(36 t, it becomes clear that stock
analysts turn to CDS markets in order to devise their future forecasts. Thus, the results
add empirical evidence to Hypothesis 3 and Hypothesis 4, assuming that the relationship between the dispersion of stock analysts’ forecasts and future CDS spreads are
positive with CDS spreads leading the dispersion of stock analysts’ forecasts. This
finding is also aligned to existing literature on the dispersion of stock analysts’ fore-
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
111
casts and future stock returns. Dieter et al. (2002), for example, show that the high dispersion of stock analysts’ forecasts is associated with lower stock returns. Again,
based on Merton (1974) lower stock returns are of course comparable to higher levels
of CDS spreads. However, applied to the data sample we do not find empirical proof
that ¨6't-l actually lead the CDS markets, and thus our results - at least on the level of
monthly average changes - contradict Diether et al. (2002) with regard to lead-lag
structures.
A third pairwise correlation can be indentified between FEPSt-l and SDt-l. The results,
displayed in Table 6-2, show that the FEPSt-l seem to be uncorrelated on a total scale
with SDt-l HJ ȡ¨)(36 t, ¨6' t-1) = 0.0024) and support the idea that no correlation
between FEPSt-l and SDt-l exists at all. However, on the level of absolute values a clear
pattern in favor of negative correlation structures can be observed HJȡ)(36t, SDt)
= - 0.3258). The results on behalf of absolute values support the idea of lead-lag structures between analysts’ forecasts and the corresponding dispersion in both direction,
whereas in case of¨)(36
t-l
and ¨6' t-1 we only find signs of ¨)(36t-l leading the
formation of ¨6't. Against the background of index affiliation correlation between
¨)(36t-l and ¨6' t-l leads to rather opposing results: CDX-linked reference entities
support positive spill-over effects between forecasts and standard deviation (e.g.
ȡ¨)(36t, ¨6' t) = 0.1216) whereas iTraxx-linked entities reveal negative correlation
VWUXFWXUHVHJȡ¨)(36t, ¨6't) = -0.1456).
FEPSt
FEPSt-1
FEPSt-2
FEPSt-3
FEPSt
FEPSt-1
FEPSt-2
FEPSt-3
ʌ;^t-l, FEPSt-l)
ѐ&W^t
ѐ&W^t-1
ѐ&W^t-2
ѐ&W^t-3
ʌ;ѐ^t-l, ѐ&W^t-l)
ʌ;^t-l, FEPSt-l)
ѐ&W^t
ѐ&W^t-1
ѐ&W^t-2
ѐ&W^t-3
ʌ;ѐ^t-l, ѐ&W^t-l)
ʌ;^t-l, FEPSt-l)
ѐ&PSt
ѐ&W^t-1
ѐ&W^t-2
ѐ&W^t-3
ʌ;ѐ^t-l, ѐ&W^t-l)
ѐ^t
-0.0253*
-0.0307*
-0.0346*
-0.0200*
CDSt
0.1553*
0.1793*
0.2055*
0.2275*
ѐ^t
0.0068
-0.0355*
-0.0529*
-0.0511*
CDSt
0.0016
0.0311
0.0582
0.0837*
ѐ^t
-0.0512*
-0.0269*
-0.0199*
0.0050
CDSt
FEPSt
0.0702*
FEPSt-1
0.0972*
FEPSt-2
0.1239*
FEPSt-3
0.1478*
*Level of significance (ɲ сϱйͿ͕ƚŚĂƚʌ тϬ
Total
iTraxx
CDX
CDSt
CDSt-1
CDSt-2
CDSt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
CDSt
CDSt-1
CDSt-2
CDSt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
CDSt
CDSt-1
CDSt-2
CDSt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
FEPSt
0.0702*
0.0410
0.0081
-0.0226
ѐ&W^t
-0.0253*
-0.0296*
-0.0480*
-0.0303*
FEPSt
0.1553*
0.1328*
0.1031*
0.0727
ѐ&W^t
0.0068
-0.0142
-0.0397*
-0.0114
FEPSt
0.0016
-0.0329
-0.0684
-0.0993*
ѐ&W^t
-0.0512*
-0.0420*
-0.0547*
-0.0454*
ʌ;^t-l, SDt-l)
ʌ;ѐ^t-l, ѐ^t-l)
ʌ;^t-l, SDt-l)
ʌ;ѐ^t-l, ѐ^t-l)
ʌ;^t-l, SDt-l)
ʌ;ѐ^t-l, ѐ^t-l)
SDt
SDt-1
SDt-2
SDt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
SDt
SDt-1
SDt-2
SDt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
SDt
SDt-1
SDt-2
SDt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
CDSt
0.1476*
0.1134*
0.0880*
0.0701*
ѐ^t
0.0586*
0.0024
-0.0169
0.0063
CDSt
0.0883*
0.0401
0.0192
0.0076
ѐ^t
0.0381*
-0.0066
-0.0173
0.0139
CDSt
0.1959*
0.1730*
0.1439*
0.1210*
ѐ^t
0.0750*
0.0097
-0.0167
-0.0002
CDSt
CDSt-1
CDSt-2
CDSt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
CDSt
CDSt-1
CDSt-2
CDSt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
CDSt
CDSt-1
CDSt-2
CDSt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
SDt
0.1476*
0.1530*
0.1656*
0.1694*
ѐ^t
0.0586*
0.0391*
0.0214*
0.0127
SDt
0.0883*
0.0964*
0.1158*
0.1219*
ѐ^t
0.0381*
0.0289*
0.0476*
0.0294*
SDt
0.1959*
0.1993*
0.2062*
0.2080*
ѐ^t
0.0750*
0.0472*
0.0002
-0.0007
ʌ;&W^t-l, SDt-l)
ʌ;ѐ&W^t-l, ѐ^t-l)
ʌ;&W^t-l, SDt-l)
ʌ;ѐ&W^t-l, ѐ^t-l)
ʌ;&W^t-l, SDt-l)
ʌ;ѐ&W^t-l, ѐ^t-l)
SDt
SDt-1
SDt-2
SDt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
SDt
SDt-1
SDt-2
SDt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
SDt
SDt-1
SDt-2
SDt-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
FEPSt
-0.3258*
-0.3103*
-0.3060*
-0.2965*
ѐ&W^t
0.0024
0.0083
-0.0082
0.0134
FEPSt
-0.4393*
-0.3731*
-0.3354*
-0.3051*
ѐ&W^t
-0.1456*
0.0844*
0.0396
0.0143
FEPSt
-0.2323*
-0.2574*
-0.2796*
-0.2867*
ѐ&W^t
0.1216*
-0.0529*
-0.0467*
0.0128
FEPSt
FEPSt-1
FEPSt-2
FEPSt-3
ѐ&W^t
ѐ&W^t-1
ѐ&W^t-2
ѐ&W^t-3
FEPSt
FEPSt-1
FEPSt-2
FEPSt-3
ѐ&W^t
ѐ&W^t-1
ѐ&W^t-2
ѐ&W^t-3
FEPSt
FEPSt-1
FEPSt-2
FEPSt-3
ѐ&W^t
ѐ&W^t-1
ѐ&W^t-2
ѐ&W^t-3
SDt
-0.3258*
-0.2739*
-0.2496*
-0.2196*
ѐ^t
0.0024
-0.0371*
-0.0299*
-0.0263*
SDt
-0.4393*
-0.3366*
-0.2824*
-0.2422*
ѐ^t
-0.1456*
-0.0396*
-0.0242*
-0.0411*
SDt
-0.2323*
-0.2212*
-0.2216*
-0.2005*
ѐ^t
0.1216*
-0.0352*
-0.0346*
-0.0144
112
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
Table 6-2: Analysis of Correlation
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
113
6.5.2 Lead-lag Structures
Panel Data Analysis
In order to apply the ensuing regression models to the reference entities, we first need
to check the individual time series of the data sample for their stationary attributes.
Only if the times series prove to be stationary (e.g. the null hypothesis stating nonstationary is rejected) are we able to include them in the subsequent regression analysis. Throughout the test for stationarity we apply three different test statistics: the
Augmented Dickey-Fuller test, the Phillips-Perron test and the Kwiatkowski-PhillipsSchmidt-Shin test. It is important to note that both the Augmented Dickey-Fuller test
and the Phillips-Perron test use a null hypothesis, stating the existence of a unit root
(non-stationary), whereas the Kwiatkowski-Phillips-Schmidt-Shin test relies on a null
hypothesis, stating the non existence of a unit root (stationary). On a 5% significance
level, the test results displayed for the time series covering absolute values of CDSi,t,
FEPSi,t and SDi,t in most cases lead to the acceptance of the hypothesis stating non stationary processes for the analyzed time series processes with i corresponding to the
individual reference entity. Whereas in the case of average monthly changes of CDSi,t,
FEPSi,t and SDi,t, the test results support the existence of stationary processes. Due to
the fact that the average FEPSi,t might be biased by the amount of outstanding stock as
well as by the different currency levels, we already placed a strong emphasize on the
average monthly ¨&'6 i,t, ¨)(36 i,t and ¨6' i,t. The results of unit root tests, crosschecking the time series of the data sample for stationarity attributes, confirm the initial decision to focus primarily on ¨&'6 i,t, ¨)(36 i,t and ¨6' i,t. For the panel data
analysis as well as VAR analysis, we therefore rely on the average changes in the CDS
spreads, stock analysts’ forecasts as well as the dispersion of the analysts’ forecasts.
Since the analysis of correlation is a commonly used approach for assessing the current levels of correlation, but not the most definite measure for evaluating lag structures, we analyze inter-temporal spill-over effects between the CDS markets and stock
analysts’ forecasts in the following, using a variety of different regression models.
Based on a fixed effect model, we first applied a panel data analysis to a data sample
of a total of 204 different reference entities for mean changes in CDSi,t-l, FEPSi,t-l and
SDi,t-l (see Table 6-3). Since the analysis of correlation already revealed different pat-
114
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
terns between the CDX and iTraxx series, it is differentiated throughout the panel data
analysis between these two index subgroups. The applied panel regression model follows a pool least squares approach and includes 58 monthly observations (March 2004
– December 2008) as well as lag structures of up to three months. Fixed effects are assumed both for cross section as well as period effects. A random effects model applied
to the data sample leads to comparable results. For each index sub group the corresponding regression output with ¨&'6¨)(36DQG¨6'UHVSHFWLYHO\DVGHSHQGHQWYariables are illustrated in Table 6-3. All regression coefficients with a level of significDQFHRIĮ DUHLQGLFDWHGZLWKDQDVWHULVNThe Durbin-Watson test statistics reveal
no signs of significant autocorrelation throughout the performed panel regressions.
Panel A of Table 6-3 summarizes the output of a panel regression within all reference
entities belonging to the CDX index. The very first regression set on the left column
shows– as already outlined in Table 6-2 – a statistically significant (negative) correlation structure on a level that is not exposed to lag structures. Thus, for the CDX index
no empirical proof is found to reject Hypothesis 1 stating that increasing stock analysts’ forecasts are associated by decreasing CDS spread levels. Even so, the results
for iTraxx-related entities do not indicate a significant positive correlation or no correlation at all we also do not find empirical proof for a significant negative correlation.
Against WKHEDFNJURXQGRIDSUHGHILQHGVLJQLILFDQFHOHYHORIĮ , the lag structures
do not appear to have any impact at all. The results of Panel A are confirmed – with
the exception of ¨)(36 t-2 on a significance level of Į 10% – for the iTraxx related
entities (Panel B). Thus, in relation to Hypothesis 2, we show that indeed, no lag structures exist between the average changes in stock analysts’ forecasts and average
changes in CDS spreads. With the lag structures playing no meaningful role, we conclude that the co-movement between ¨&'6 t-l and ¨)(36 t-1 exists and takes place instantly – specifically in the case of CDX related entities (Panel A). We do not find any
empirical proof that the ¨6't-l time series lead the ¨&'6t time series. However, the
coefficient referring to ¨6' t is significant on the Į 10% level and therefore supports
Hypothesis 3 together with the coefficient of ¨&'6 t (Į 10%) in the panel regression
with ¨6't as a dependent variable (e.g. regression coefficients of ¨&'6t for the total
Coefficient
0.0575*
0.0002
0.0006
0.0014
0.0001
0.0013
0.0011
0.0002
0.0005
0.4319
2.0906
Dependent Variable: ѐ^t
Reference Entities included: 204
Total Panel Observations: 11,832
Dependent Variable: ѐ^t
Reference Entities included: 91
Total Panel Observations: 5,278
Variable
Coefficient
C
0.0597*
ѐ&W^t
0.0010
ѐ&W^t-1
0.0010
ѐ&W^t-2
0.0016
ѐ&W^t-3
0.0002
ѐ^t
0.0000
ѐ^t-1
-0.0011
ѐ^t-2
-0.0012
ѐ^t-3
0.0010
R-squared
0.5621
Durbin-Watson stat
2.0811
Dependent Variable: ѐ^t
Reference Entities included: 113,
Total Panel Observations: 6,554
Variable
Coefficient
C
0.0553*
ѐ&W^t
-0.0158*
ѐ&W^t-1
0.0028
ѐ&W^t-2
0.0012
ѐ&W^t-3
0.0001
ѐ^t
0.0030*
ѐ^t-1
0.0023
ѐ^t-2
0.0016
ѐ^t-3
-0.0006
R-squared
0.4027
Durbin-Watson stat
2.0780
Variable
C
ѐ&W^t
ѐ&W^t-1
ѐ&W^t-2
ѐ&W^t-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
R-squared
Durbin-Watson stat
Ύ>ĞǀĞůŽĨƐŝŐŶŝĨŝĐĂŶĐĞ;ɲсϱйͿ
Panel C (Total)
Panel B (iTraxx)
Panel A (CDX)
Std. Error
0.0021
0.0010
0.0010
0.0010
0.0002
0.0007
0.0008
0.0009
0.0009
F-statistic
Prob(F-ƐƚĂƚŝƐƚŝĐͿ
Std. Error
0.0029
0.0009
0.0009
0.0009
0.0002
0.0008
0.0012
0.0013
0.0013
F-statistic
Prob(F-ƐƚĂƚŝƐƚŝĐͿ
Std. Error
0.0029
0.0057
0.0056
0.0054
0.0052
0.0011
0.0012
0.0013
0.0013
F-statistic
Prob(F-ƐƚĂƚŝƐƚŝĐͿ
Prob.
0.0000
0.8145
0.5294
0.1563
0.6469
0.0577
0.2103
0.8630
0.5769
3.2804
0.0000
Prob.
0.0000
0.2661
0.3065
0.0820
0.4013
0.9405
0.3465
0.3577
0.4293
4.2424
0.0000
Prob.
0.0000
0.0060
0.6089
0.8187
0.9915
0.0085
0.0574
0.2090
0.6233
2.4290
0.0000
Variable
C
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
R-squared
Durbin-Watson stat
Coefficient
0.0467*
0.0073
-0.0611
-0.0414
0.0650
-0.0234*
0.0716*
0.0600*
0.0089
0.0381
2.1045
Dependent Variable: ѐ&W^t
Reference Entities included: 204
Total Panel Observations: 11,832
Dependent Variable: ѐ&W^t
Reference Entities included: 91
Total Panel Observations: 5,278
Variable
Coefficient
C
0.0909
ѐ^t
0.1938
ѐ^t-1
-0.0230
ѐ^t-2
-0.0698
ѐ^t-3
0.1753
ѐ^t
-0.0280*
ѐ^t-1
0.2431*
ѐ^t-2
0.0655*
ѐ^t-3
-0.0195
R-squared
0.0701
Durbin-Watson stat
2.0962
Dependent Variable: ѐ&W^t
Reference Entities included: 113
Total Panel Observations: 6,554
Variable
Coefficient
C
0.0122
ѐ^t
-0.0737*
ѐ^t-1
-0.0135
ѐ^t-2
0.0440
ѐ^t-3
-0.0319
ѐ^t
-0.0115*
ѐ^t-1
-0.0609*
ѐ^t-2
0.0541*
ѐ^t-3
-0.0031
R-squared
0.1658
Durbin-Watson stat
2.0478
Std. Error
0.0227
0.0877
0.0882
0.0897
0.0928
0.0066
0.0080
0.0086
0.0086
F-statistic
Prob(F-ƐƚĂƚŝƐƚŝĐͿ
Std. Error
0.0504
0.2092
0.2107
0.2133
0.2221
0.0121
0.0175
0.0193
0.0193
F-statistic
Prob(F-ƐƚĂƚŝƐƚŝĐͿ
Std. Error
0.0070
0.0275
0.0275
0.0281
0.0288
0.0025
0.0025
0.0027
0.0026
F-statistic
Prob(F-ƐƚĂƚŝƐƚŝĐͿ
Prob.
0.0392
0.9335
0.4888
06447
0.4834
0.0004
0.0000
0.0000
0.3003
1.7122
0.0000
Prob.
0.0714
0.3543
0.9132
0.7436
0.4300
0.0208
0.0000
0.0007
0.3125
2.4922
0.0000
Prob.
0.0833
0.0073
0.6235
0.1165
0.2692
0.0000
0.0000
0.0000
0.2362
7.1596
0.0000
Variable
C
ѐ^t
ѐ^t-1
ѐ^t-2
ѐ^t-3
ѐ&W^t
ѐ&W^t-1
ѐ&W^t-2
ѐ&W^t-3
R-squared
Durbin-Watson stat
Coefficient
0.2078*
0.2350
0.0294
0.0794
0.0923
-0.0519*
-0.0068
-0.0058
-0.0001
0.0317
1.7542
Dependent Variable: ѐ^t
Reference Entities included: 204
Total Panel Observations: 11,832
Dependent Variable: ѐ^t
Reference Entities included: 91
Total Panel Observations: 5,278
Variable
Coefficient
C
0.2540*
ѐ^t
0.0061
ѐ^t-1
-0.5987*
ѐ^t-2
0.2305
ѐ^t-3
-0,3944
ѐ&W^t
-0.0416*
ѐ&W^t-1
0.0061
ѐ&W^t-2
-0.0051
ѐ&W^t-3
-0.0001
R-squared
0.0315
Durbin-Watson stat
1.5371
Dependent Variable: ѐ^t
Reference Entities included: 113
Total Panel Observations: 6,554
Variable
Coefficient
C
0.1750
ѐ^t
0.3614*
ѐ^t-1
0.4425*
ѐ^t-2
0.0711
ѐ^t-3
0.4860*
ѐ&W^t
-0.2314*
ѐ&W^t-1
0.0158
ѐ&W^t-2
-0.0379
ѐ&W^t-3
0.0411
R-squared
0.0620
Durbin-Watson stat
2.1452
Std. Error
0.0319
0.1244
0.1250
0.1271
0.1316
0.0131
0.0131
0.0131
0.0031
F-statistic
Prob(F-ƐƚĂƚŝƐƚŝĐͿ
Std. Error
0.0578
0.2414
0.2430
0.2460
0.2561
0.0158
0.0158
0.0159
0.0037
F-statistic
Prob(F-ƐƚĂƚŝƐƚŝĐͿ
Std. Error
0.0350
0.1380
0.1383
0.1409
0.1450
0.0586
0.0569
0.0570
0.0570
F-statistic
Prob(F-ƐƚĂƚŝƐƚŝĐͿ
Prob.
0.0000
0.0590
0.8140
0.5323
0.4830
0.0001
0.6015
0.6580
0.9719
1.4108
0.0000
Prob.
0.0000
0.9799
0.0138
0.3488
0.1236
0.0085
0.6967
0.7464
0.9695
1.0756
0.2496
Prob.
0.0000
0.0089
0.0014
0.6138
0.0008
0.0001
0.7821
0.5066
0.4711
2.3800
0.0000
Table 3 illustrates the results of a panel regression based on a fixed effects model. Panel A ;ͿƌĞƉƌĞƐĞŶƚƐĂůůƌĞĨĞƌĞŶĐĞĞŶƚŝƚŝĞƐǁŝƚŚĂĨĨŝůŝĂƚŝŽŶƚŽƚŚĞy;ŝdƌĂdždžͿŝŶĚĞdžĨĂŵŝůLJ͘WĂŶĞůconsist of all 204 reference entities.
The applied panel regression model follows the pool least squares method. After adjusting for lag structures, each panel regression includes 58 observations (March 2004 - ĞĐĞŵďĞƌϮϬϬϴͿ͘ Each panel, in turn, displays
the results of the applied fixed effect regression model with changing dependent variables (ѐCDSt, ѐFEPSt, ѐSDtͿĂŶĚůĂŐƐƚƌƵĐƚƵƌĞƐů;-1, -2, -ϯͿ͘
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
Table 6-3: Panel Data Analysis (Fixed Effect Model)
115
116
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
sample is 0.2350). The panel regression model with ¨6' t as a dependent variable also
reveals significance levels of the ¨&'6t-l coefficients (e.g. the regression coefficient of
¨&'6t-1 in Panel B is -0,5987) that allows us to conclude that the CDS markets lead
the formation of stock analysts’ dispersion (Hypothesis 4). The results of the preceding
analysis of correlation are thus confirmed. With regard to index affiliation, we again
observe different patterns between Panel A and Panel B. Whereas in the case of Panel
A lead-lag relationships become the strongest at a level of three months, this is true for
Panel B on the level of lag of one month. Finally, we also show that the ex-ante dispersion (¨6't-l) leads the formation of the stock analysts’ future estimations (¨)(36t).
Vector Auto Regression Analysis
The application of panel data analysis allows us to compare co-movement within different subgroups on a combined level. In the following we perform a VAR analysis for
each reference entity, based on Norden and Weber (1997) and Engsted and Tanggard
(2004), in order to detect lead-lag relationships on a firm specific level. VAR models
are used in order to explore forecast properties between interrelated time series under
consideration of lag structures. More specifically, each dependent variable is modeled
on the basis of the lagged values of all other variables. Following Norden and Weber
(1997) we define a VAR model as:
'CDSt
'FEPSt
'SDt
L
L
L
l 1
l 1
l 1
c1 ¦ E1l 'CDSt l ¦ J 1l 'FEPSt l ¦ G1l 'SDt l H1t
L
L
L
l 1
l 1
l 1
c2 ¦ E 2l 'CDSt l ¦ J 2l 'FEPSt l ¦ G 2l 'SDt l H 2t
L
L
L
l 1
l 1
l 1
c3 ¦ E3l 'CDSt l ¦ J 3l 'FEPSt l ¦ G 3l 'SDt l H 3t
Formula (1) - (3) represent the different VAR models in order to identify lead-lag
structures between the dependent variables ¨&'6 t, ¨)(36 t and ¨6't and the lagged
values ¨&'6t-l, ¨)(36t-l and ¨6't-l. In this context l indicates the individual time lags
and ranges from 1 to 3 on a monthly level. Eil , J il and G il are the individual estimation
coefficients of the dependent variables and H it stands for innovations (error term) uncorrelated with all dependent variables as well with the constants ci. Note that the
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
117
VAR models introduced above are individually applied to each reference entity in the
ensuing VAR estimation.
Table 6-4 illustrates the results sorted according to the subgroups CDX, iTraxx as well
as the total sample. After adjusting for lag structures each VAR estimate includes 58
months (March 2004 - December 2008). Displayed coefficients are mean values computed on the basis of entity specific VAR outputs. We thus computed 204 different
VAR estimations for each dependent variable, and based on these firm specific results
we calculated mean values on behalf of CDX, iTraxx as well as the total data sample
as displayed in Table 6-4. Column (1) and (2) report the number of reference entities
IRU ZKLFK WKH UHVXOWVSURYH WREH VLJQLILFDQW ZLWK Į DQGĮ UHVSectively.
We again use ¨&'6 t-l, ¨)(36 t-l and ¨6' t-l both as dependent and independent variables in order to meet the condition of stationary input factors. Overall, the VAR approach confirms the findings obtained throughout the preceding panel data analysis.
Starting with the set of VARs computed for¨&'6
t
as the dependent variable, inter-
temporal relationships are most likely to occur within the¨&'6
t-l
time series. For 56
reference entities – or 27.45% of the total sample – we detect the inter-temporal linkages between ¨&'6 t and ¨&'6 t-1 Į :LWKDPD[LPXPRIUHIHUHQFHHQWities (in the case of ¨)(36 t-3) the lagged co-movement between ¨&'6 t and ¨)(36t-l is
significantly lower. This pattern is also confirmed for the VAR model using ¨)(36 t as
the dependent variable (at 18 entities in the case of¨&'6
t-1).
Thus, the test statistics
do not allow us to reject Hypothesis 2. Based on a VAR approach empirical proof of
the existence of the inter-temporal relationships between¨&'6
t-l
and ¨)(36 t-l time
series in less than 10% of the reference entities is found. It is therefore rather unlikely
that either CDS markets or stock analysts’ forecasts will lead the other in each case.
However, since we plot for each month mean estimates and CDS spreads observed on
the day I/B/E/S published the mean estimates, it is rather comprehensible that - if existing at all - potential adjustment trends are already reflected in CDS spreads throughout a monthly analysis. In addition, we observe that the variables, on average, are most
frequently interrelated with their own lagged values. The only exception is found for
¨6't, where the ¨&'6t-l time series reaches a comparably high number of inter-
118
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
Table 6-4: Vector Auto Regression Analysis (Lag Structures)
CDX
Dependent
Undependent
C
ѐDSt-1
ѐ^t-2
ѐDSt-3
ѐ&EPSt-1
ѐ&W^t-2
ѐ&EPSt-3
ѐ^t-1
ѐ^Dt-2
ѐ^Dt-3
ѐ^t
# of entities: 113
ѐ&W^t
# of entities: 113
ѐ^t
# of entities: 113
Coeff.
;ϭͿ
;ϮͿ
Coeff.
;ϭͿ
;ϮͿ
Coeff.
;ϭͿ
;ϮͿ
0.0553
0.1531
-0.1273
0.0587
-0.1555
-0.1078
-0.0592
-0.0024
-0.0025
-0.0032
6
19
15
5
6
3
5
7
8
10
19
31
20
9
8
5
11
10
14
15
0.0063
-0.0128
0.0162
0.0049
-0.0075
-0.0158
0.0606
-0.0077
-0.0017
-0.0035
23
7
6
6
8
8
22
10
7
14
37
11
12
9
13
11
27
13
7
19
0.2593
0.4037
-0.0520
0.1802
-0.3925
-0.0999
0.1783
-0.1175
-0.1139
-0.0975
9
14
9
13
3
1
1
6
8
7
30
25
13
17
3
3
6
12
14
13
iTraxx
Dependent
Undependent
C
ѐDSt-1
ѐDSt-2
ѐDSt-3
ѐ&EPSt-1
ѐ&EPSt-2
ѐ&W^t-3
ѐ^Dt-1
ѐ^Dt-2
ѐ^t-3
ѐ^t
# of entities: 91
ѐ&W^t
# of entities: 91
ѐ^t
# of entities: 91
Coeff.
;ϭͿ
;ϮͿ
Coeff.
;ϭͿ
;ϮͿ
Coeff.
;ϭͿ
;ϮͿ
0.0633
0.1153
-0.1211
0.1203
-0.1487
-0.3945
-0.1391
0.0062
0.0040
0.0444
9
21
11
13
1
4
2
2
4
7
19
25
18
21
3
8
7
7
9
9
0.0283
0.0540
0.0600
0.1243
-0.0194
-0.1123
-0.0635
0.1658
0.0289
-0.0046
20
1
2
3
8
4
11
16
9
12
25
5
6
3
12
8
13
20
13
13
0.2056
-0.0156
0.1994
-0.0868
-0.2161
-0.0141
-0.1489
-0.0781
-0.0432
-0.0429
4
7
8
7
9
4
3
9
9
8
10
9
8
10
12
6
7
16
10
14
;ϭͿ
13
21
17
20
12
5
4
15
17
15
;ϮͿ
40
34
21
27
15
9
13
28
24
27
Total
Dependent
Undependent
C
ѐDSt-1
ѐDSt-2
ѐDSt-3
ѐ&W^t-1
ѐ&EPSt-2
ѐ&EPSt-3
ѐ^Dt-1
ѐ^Dt-2
ѐ^Dt-3
ѐ^t
# of entities: 204
Coeff.
0.0588
0.1365
-0.1246
0.0857
-0.1525
-0.2337
-0.0943
0.0014
0.0003
0.0177
;ϭͿ
15
40
26
18
7
7
7
9
12
17
ѐ&W^t
# of entities: 204
;ϮͿ
38
56
38
30
11
13
18
17
23
24
Coeff.
0.0160
0.0168
0.0354
0.0573
-0.0127
-0.0582
0.0061
0.0685
0.0117
-0.0040
ѐ^t
# of entities: 204
;ϭͿ
42
8
8
9
16
12
33
26
16
26
;ϮͿ
63
16
18
12
25
19
40
33
20
32
Coeff.
0.2357
0.2333
0.0584
0.0630
-0.3151
-0.0498
0.0347
-0.1002
-0.0829
-0.0735
temporal linkages and in the case lag -1 even outperforms the absolute number of
lagged co-movements between ¨6' t and ¨6't-1 (34 versus 28 detected interrelations).
The pattern of inter-temporal dynamics between CDS spreads and the dispersion of the
stock analysts’ forecasts, as documented in Table 6-3 also hold for the application of a
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
119
VAR model. Since the inter-temporal linkages are higher in the case of¨6' t, being a
function of the lagged values of¨&'6
t-l,
we conclude that changes in CDS spreads
lead to changes in SD. In other words, the empirical outcome is assessed as additional
proof for Hypothesis 4. Reviewing the VAR outcomes on behalf of¨)(36
t
functions
of lagged values, we not only notice the aforementioned inter-temporal linkages with
their own lagged times series ¨)(36 t-l, but we also note the intense co-movement with
¨6'-t. Accordingly, we consider the dispersion of the stock analysts’ forecasts to lead
to future changes in the forecasts.
However, after cross-checking for index affiliation, we document slightly different
patterns for the CDX and iTraxx index members respectively. The lagged comovement between ¨6' t and ¨&'6 t-l is higher for reference entities of the CDX index, whereas the inter-temporal relationship between¨)(36
t
and ¨6' t-l is stronger
against the background of the reference entities affiliated to the iTraxx index.
6.5.3 Long-run Equilibrium Relationship
In the previous chapter we focused on lead-lag dynamics between changes in CDS
spreads, changes in mean stock analysts’ forecasts as well as changes in corresponding
stock analysts’ uncertainty with regard to future company earnings. Both panel data
analysis as well as VAR estimation (Table 6-3 and 6-4) relied on ¨&'6¨)(36DQG
¨6'LQRUGHUWRPHHWWKHPHWKRGLFDOUHTXLUHPHQWRI stationary input variables. Application of cointegration techniques in turn allows us to incorporate variables with nonstationary attributes: If non-stationary time series can be aggregated into a stationary
linear combination, we label such a coherence as a cointegration equation. Empirical
proof of cointegrating time series can be interpreted as a long-run equilibrium relationship among the analyzed non-stationary variables. Long-run equilibrium in turn is additional prove of co-movement dynamics between two time series. Since nonstationary attributes were initially detected in the case of CDS spreads, FEPS and SD
time series of absolute values, we test for cointegration between these three time series
for each individual reference entity in the following. Relying on a significance level of
Į DQGĮ UHVSHFWLYHO\we decide for each reference entity of the data sample if CDS spreads, FEPS and SD are cointegrated between each other. The applied
methodology relies on a cointegration test developed by Johansen (1991, 1995), which
120
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
has been repeatedly used in the context of co-movement between CDS spreads and adjacent asset classes (see for example Blanco et al., 2005, who applied a Johansen cointegration test for co-movement between CDS spreads and corporate bond yields).
More specifically, in order to decide whether cointegration is present or not, we rely
upon an unrestricted cointegration rank trace test and use MacKinnon et al. (1999) p values to come up with the above mentioned significance levHOV RI Į DQG Į 10%.
Against this background Table 6-5 reports the results of Johansen (1991, 1995) cointegration test applied to our data sample of 204 reference entity and the underlying individual time series of CDS spreads, FEPS and SD for the time period March until December 2008. In detail, the performed Johansen (1991, 1995) cointegration test follows
a linear deterministic trend. Lags of the first reference difference terms relate to a lag
of up to 2. Thus, lag-structures are also taken into account. In line with previous analysis, we again divide our data sample according to affiliation to the two index series
CDX and iTraxx, and provide the results on an aggregated level for the total sample.
Given three different time series for each reference entity (CDS spreads, FEPS and
SD) we come up with three different test settings for each reference or a total of 612
performed cointegration tests. Throughout the three different test settings for cointegration we again encounter different outcomes for CDX and iTraxx index series.
As we assume with regard to Hypothesis 1 that CDS spreads are negatively correlated
with stock analysts’ forecasts and already found first proof throughout the correlation
analysis as well as the panel regression we expect the existence of long-run equilibrium between CDS and FEPS time series. In total, cointegration between CDS and
)(36WLPHVHULHVSURYHVWRH[LVWIRUUHIHUHQFHHQWLWLHVĮ RU7KXV
empirical support for Hypothesis 1 is limited on a total sample level. However, differentiation between iTraxx and CDX series reveals higher empirical support for iTraxxlinked reference entities with 46.RIFRLQWHJUDWHGWLPHVHULHVĮ 7KLVLVSDrticular interesting given the fact that throughout the panel data analysis (Table 6-3) we
documented rather opposing results with regard to significance levels between¨&'6
and ¨)(36HJVLJQLILFDQWQHJDWLYHFRUUHODWLRQLQWKHFDVHRI&';EXWQRWLQWKHFDVH
of iTraxx). Long-run dynamics are obviously not affected by direct correlation struc-
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
121
tures. Thus, deviation in the short-term does not affect the existence of a long-run
equilibrium. Against this background short-term deviation can be viewed as a kind of
permanent adjustment process towards long-run stability (Norden and Weber, 2007).
The second set of cointegration tests reported in Table 6-5 corresponds to CDS and SD
time series. In direct comparison to the test results of CDS spreads and FEPS time series, we observe that the total number of identified firm-specific time series for CDS
spreads and dispersion of stock analysts’ forecasts are higher for both the CDX- and
the iTraxx-linked series (e.g. at a significance level of 0.10 on average 63.2% of analyzed reference entities correspond to cointegrated CDS and SD time series). Overall,
these high levels can be interpreted as empirical support of H3, which posits a correlation between CDS and SD. In the case of the CDX index this figure is even higher,
with 78 entities or 69.0% of the total subgroup sample, and confirms the patterns already documented in Table 6-4 with respect to index affiliation: VAR models applied
to the CDX sample detected a high number of lead-lag structures with CDS spreads
leading SD formation. Besides identifying existing lead-lag structures between CDS
and SD we thus also document a long-run equilibrium relationship.
Finally, the highest level of cointegration is documented for the total data sample between the time series of FEPS and SD with 58.88% Į DQG.Į or 119 and 167 cases of existing cointegrating equations. This finding of course is not
unexpected given the fact that dispersion of stock analysts is directly derived from
stock analysts’ forecasts. Therefore, it is reasonable to argue that both time series will
follow a long-run relationship of equilibrium. Comparison of CDX and iTraxx-linked
results shows that the percentage of CDX affiliated reference entities to be cointegrated is higher than those observed in the iTraxx-based sample (e.g. 80.5% versus
71.4% on a 0.10 significance level).
Comparison of our results with existing research on cointegration attributes of CDS
spreads with corresponding corporate bonds yields confirm our empirical outcomes
with regard to stock analysts’ forecasts. Both Blanco et al. (2005) as well as Norden
and Weber (2007) document comparable levels of cointegration in the case of bond
and CDS spreads to the ones we obtained through our analysis on the cointegration of
122
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
Table 6-5: Long-run Equilibrium Relationship between CDS Spreads and Stock
Analysts’ Forecasts
Cointegration of CDS and FEPS Times Series
ɲсϱйΎ
# of entities
CDX
24
iTRaxx
26
Total
50
ŝŶйŽĨŐƌŽƵƉƐĂŵƉůĞ
21.24
28.57
24.51
ɲсϭϬйΎ
# of entities
35
42
77
ŝŶйŽĨŐƌŽƵƉƐĂŵƉůĞ
30.97
46.15
37.75
Cointegration of CDS and SD Time Series
ɲсϱйΎ
# of entities
CDX
67
iTraxx
38
Total
105
ŝŶйŽĨƚŽƚĂůƐĂŵƉůĞ
59.29
41.76
51.47
ɲсϭϬйΎ
# of entities
78
51
129
ŝŶйŽĨŐƌŽƵƉƐĂŵƉůĞ
69.03
56.04
63.24
Cointegration of FEPS and SD Time Series
ɲсϱйΎ
# of entities
CDX
69
iTRaxx
50
Total
119
ŝŶйŽĨƚŽƚĂůƐĂŵƉůĞ
61.06
54.95
58.33
ɲсϭϬйΎ
# of entities
91
65
167
ŝŶйŽĨŐƌŽƵƉƐĂŵƉůĞ
80.53
71.43
81.86
*based on MacKinnon-Haug-DŝĐŚĞůŝƐ;ϭϵϵϵͿƉ-values
CDS spreads and stock analysts’ forecasts and the underlying dispersion SD. Both studies also propose different cointegration patterns with respect to Europe and US assets.
6.6 Conclusion
This chapter aims to analyze co-movement and lead-lag structures between stock analysts’ earnings forecasts and CDS spreads. First, from the perspective of CDS markets
we assess the importance of - and interaction with - other information agents than rating agencies. Second, from the angle of stock analysts, we investigate if their guidance
attributes are limited to stock markets only or if they also impact pricing processes on
adjacent capital markets. By doing so, we also address the issue of information distribution between CDS markets and stock analysts’ forecasts. Based on the empirical
findings (chapter 6.5) we are able to assess to what extend information asymmetries
between CDS spreads (as a dummy for CDS markets) and analysts’ forecasts (as a
dummy for stock markets) exist. Throughout the empirical part we approach spill-over
effects between stock analysts’ forecasts and CDS spreads on two different levels:
mean stock analysts’ earnings forecasts and dispersion of mean stock analysts’ earnings forecasts. Each corresponding time series is matched with the corresponding CDS
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
123
spreads. As a kind of robustness check we also control for the interacting dynamics
between mean forecasts and its’ dispersion.
Against this background our main findings are threefold: First, we find significant
signs of non-lagged co-movement between CDS spreads and mean earnings forecasts.
We show that higher credit spreads are associated with lower mean forecasts as well as
higher levels of forecast dispersion. Second, we observe significant lead-lag structures
between the dispersion of stock analysts’ forecasts and CDS spreads with the latter
leading the first. These lead-lag dynamics are confirmed by a long-run equilibrium relationship taking place between forecast dispersion and CDS spreads. Third, our empirical results indicate significantly different patterns between reference entities linked
to the U.S. or Europe.
Referring to the detected non-lagged co-movement dynamics between CDS spreads
and analysts’ forecasts (see chapter 6.5.1) our results indicate that specifically in the
case the CDX index both CDS markets and stock analysts not only tend to rely on the
same information sources but also seem to process the content of information simultaneously. Based on these findings we additionally argue that none of the two has access
to exclusive information. Assuming a simple asset-based credit risk model (e.g. Merton, 1974) the direction of detected correlation structures is also from an economic
perspective rather reasonable: Decreasing future estimates lower the reference entity’s
value. A lower entity value correlates with higher risk levels of the underlying debt or
higher CDS spreads respectively. However, we have to take into account, that we do
not observe comparatively significant patterns for the iTraxx-related entities.
A second relationship analyzed throughout this chapter in more detail refers to the dispersion of stock analyst’ forecasts and CDS spreads. We document a positive correlation structure between dispersion of mean forecasts and CDS spreads and interpret
dispersion in this context as a measure of uncertainty. Higher CDS spreads are thus
associated with higher uncertainty among stock analysts. According to the credit risk
framework of Merton (1974) risk is foremost a function of the underlying assets’ volatility. The asset’s value in turn is derived from the future earnings perspectives of the
corresponding reference entity. Thus, future earnings perspectives can be viewed as
stock analysts’ forecasts. Uncertainty about future earnings perspectives (e.g. uncer-
124
Spill-over Effects between Stock Analysts’ Earnings Forecasts and CDS Spreads
tainty among analysts) in turn affects the asset’s value volatility as well as the credit
risk level. Thus, risk is ultimately linked to the dispersion of stock analysts’ forecasts
or in other words dispersion of forecasts can be viewed as a proxy for credit risk.
Co-movement between these two adjacent market segments also shows that stock analysts are perceived as information agents on CDS markets by investors and information distribution is not bounded by different asset classes. However, since we actually
observe CDS spreads leading the dispersion of analysts’ forecasts, we note that with
regard to future earnings’ uncertainty information asymmetries exist at least to some
extend between CDS markets and stock markets (e.g. represented by stock analysts).
In addition to co-movement and lead-lag structures we also focus on long-run equilibrium relationships between (dispersion) of stock analysts’ forecasts and CDS spreads.
Against the background of a Johansen cointegration test statistic we detect significant
results confirming the existence of a long-run equilibrium relationship between dispersion of stock analysts’ forecasts and CDS spreads. Obviously the pattern for spill-over
effects between dispersion of forecasts and CDS spreads documented with regard to
correlation coefficients as well as monthly lag structures do also hold for the long run.
Throughout the empirical analysis we repeatedly observe intensifying or even opposing patterns with respect to the index affiliation (CDX or iTraxx) of the reference entities. In particular we find significant negative correlation structures between analysts’
forecasts and CDS spreads in case of reference entities headquartered in the U.S. rather than observed in case of European entities. Stronger spill-over effects might be a
function of the underlying capital market regimes with the U.S.-American one being
more integrated than European markets. Specifically, the role of analysts performed in
the U.S. can be regarded to be more prominent against the background of guidance
attributes than in Europe. Different price patterns on European and CDS markets – as
discovered throughout the empirical part of this chapter – are also confirmed by exiting literature covering spill-over effects of CDS markets with other exogenous factors.
Lehnert and Neske (2006) for example display diverging empirical results between
European and US reference entities in case of co-movement between rating adjustments and CDS spreads.
Conclusion and Outlook
125
7 Conclusion and Outlook
7.1 Summary of the Results
In the wave of the worldwide financial crisis that started in 2007/08, the functionality
of secondary credit markets has been widely criticized both by academics and practitioners. In this context, trust and confidence were frequently quoted as the key factors,
which the financial markets have failed to fulfill. Trust and confidence, in turn, are a
direct function of information asymmetries and market transparency. However, a detailed empirical analysis of the specific areas in which information asymmetries may
occur in secondary credit markets has only taken place on a rather limited scale. This
dissertation, therefore, aims to discuss the empirical evidence of information asymmetries in secondary credit markets and addresses this issue through three different empirical settings (chapters 4 to 6) by specifically focusing on the CDO and CDS markets. Against the background of secondary credit markets, the dissertation seeks to answer to what extent information asymmetries exist, how information asymmetries impact pricing structures and whether information asymmetries also exist between the
secondary credit markets and adjacent financial markets.
First, in the course of chapter 4, it is analyzed to what extent information asymmetries,
emerging from the specific characteristic of the CDO rating market, lead to rating
model arbitrage in CDO markets and thus impact the behavior of market participants.
Second, chapter 5 provides us with a detailed analysis of how different levels of information distribution on secondary credit markets (represented by the number of outstanding tranche ratings) affect the corresponding credit spreads. Third, the dissertation also targets the issue of information asymmetries regarding the inter-market perspective between the CDS markets and stock analysts’ forecasts (chapter 6).
The insights gained throughout the empirical analysis of chapters 4 to 6 can be summarized as follows:
126
Conclusion and Outlook
(I) Rating Model Arbitrage on CDO Markets: An Empirical Analysis
It is analyzed whether information asymmetry between issuers and investors leads to
rating model arbitrage in CDO markets. Rating model arbitrage is defined as the issuer’s deliberate capitalization of information asymmetry at the investor’s cost on the
basis of different rating processes. Using data from CDO transactions grouped by both
rating agencies and underlying rating methodologies, it is tested for homogeneity of
characteristic transaction features within the group and heterogeneity between the different groups. The hypothesis stating the non-existence of rating model arbitrage on
the basis of information asymmetry does not hold, as individual patterns of transaction
characteristics within each group can be identified.
Thus, empirical evidence is provided, that information asymmetry, with regard to the
rating process of secondary credit market instruments (CDO transactions) may exist
and determine market functionality as well as the behavior of the market participants
(descriptive research objective): issuers deliberately benefit from their controlling position throughout the CDO rating process and thus leverage upon existing information
asymmetries within the relationship triangle “issuer - investor - rating agencies”. Regarding CDO transactions, the applied market standard therefore systematically adds
to a specific behavioral pattern on the issuer’s side. Given the framework of rating
model arbitrage, issuers steer the rating process consequently to their own benefit,
which in turn corresponds to the publication of favorable rating outcomes only lowering credit spreads subsequently.
In more detail, the results reveal that the specific characteristics of a CDO transaction
(e.g. currency and maturity) incorporate higher explanatory power throughout the empirical section, as observed for other variables. From an economic perspective, these
findings can be explained by the specific attributes of the different underlying rating
models being particularly sensitive to selected tranche characteristics. In line with
Fender and Kiff (2005), it is argued that the impact of the rating outcome with regard
to the different rating models even varies within the seniority structure of one CDO
transaction.
On the basis of information asymmetry, it is suggested that the issuers of CDO transactions have economic incentives to take advantage of the uneven information distri-
Conclusion and Outlook
127
bution between issuers and investors and to perform rating model arbitrage. Thus defined, the issuer’s rationale is comprehensible, whereas it is questionable why investors accept their underrepresented position throughout the rating process. In terms of
the defined principal-agent relationship (chapter 2.3), it appears that the principal’s position is stronger accentuated in the case of the relationship between the issuer and the
rating agency, as observed in turn on behalf of the relationship between the investor
and the rating agency. Taking into account the theoretically strong position of the investor as a principal (e.g. the investor controls capital allocation) throughout the basic
relationship with the issuer (agent), it is even more remarkable that the investor does
not demand the issuer to at least make the dialog with the rating agencies public. However, at this point, it has to be taken into account that the CDO markets are still at a
relatively early evolutionary stage and that the investors have yet not teamed up in order to align their interests unitarily. In order to force the issuers to change the market
standards, a joint position as well as combined market power of the investors should
therefore be rather beneficial.
In addition to the evidence of rating model arbitrage, chapter 4, reveals in more detail
the specific patterns with regard to the applied rating methodologies. Since S&P and
Fitch apply the same methodology, it is particularly interesting that consistent patterns
are documented between these two rating agencies. Obviously, the rating methodologies impact on the rating outcome. This structural impact factor, in turn, sheds light on
the issue which rating agency investors should demand the issuer to obtain a rating
from. If the issuer, for example, assigns Fitch and S&P, the investor might fall victim
to a methodology-biased rating outcome. In order to avoid such biased-diluted rating
outcomes, investors should consequentially demand one rating of each rating methodology. This choice of course is directly related to monitoring costs (e.g. each additional rating is an expense factor) and should always be analyzed in close connection
with the underlying spreads. The underlying credit spreads in relation to the number of
outstanding CDO ratings are explored in more detail throughout chapter 5.
128
Conclusion and Outlook
(II) Impact of Multiple CDO Ratings on Credit Spreads
It is analyzed whether multiple ratings for CDO tranches have an impact on credit
spreads and various effects are examined with regard to the number of rating agencies
involved. Based on a data set of more than 5,000 CDO tranches, index-adjusted credit
spreads were calculated to isolate the specific credit risk per CDO tranche. First, a
negative correlation between number of ratings and credit spreads per CDO tranche is
detected, i.e. additional ratings are accompanied by lower credit spreads. Additionally,
on the basis of a valuation model, the analysis shows that multiple ratings are a significant pricing factor and interpret that investors demand an extra risk premium due to
information asymmetries between the CDO issuers and investors. Any additional rating reveals incremental information to the market and increases transparency with regard to the underlying portfolio’s credit risk. Second, however, no empirical support is
found for the hypothesis stating that marginal tranche spread reduction decreases when
additional rating agencies are added. Third, evidence is found that second or third ratings by Fitch, on average, are higher when directly compared with Moody’s and/or
S&P’s ratings per CDO tranche. This finding is in line with the existing literature on
corporate bonds and induces a bias also on CDO ratings due to their solicited character.
Throughout the empirical analysis in chapter 5, the concept of adjusted credit spreads
is introduced to separate the idiosyncratic credit risk and isolate it from potential dilution triggered by systematic risk factors. The application of index-adjusted credit
spread reduced the impact of variables linked to the tranche’s credit quality and led to
a further increase of observed influence levels of multiple ratings. Up to now, the financial literature has relied on unadjusted credit spreads only. Thus, the approach of
index-adjusted credit spreads represents a contribution in the field of applied credit
risk models.
With regard to changes in credit spread reduction in the case of moving from single to
double and double to triple ratings respectively, the findings are less distinctive. Although decreasing levels of credit spreads are displayed in all cases, the level of spread
reduction leads to rather opposing results. The hypothesis of marginal utility, which
suggests decreasing levels of spread reduction, received as much empirical support as
Conclusion and Outlook
129
the hypothesis based on a selection bias (e.g. increasing spread reduction). Additional
ratings always come along with additional costs; thus, the incremental value of additional ratings through spread reduction should at least amount to the level of costs associated with an additional rating. CDO rating costs are expected to be in the range of
4.5 bps of the underlying tranche volume. However, in both cases (single to double as
well as double to triple ratings), documented spread reduction is, on average, always
higher and thus the issuer benefits from additional ratings. The issuer seeks additional
ratings, if he believes that additional ratings are a convincing signaling instrument in
order to reduce information asymmetries, guarantee a successful placement of the
CDO tranches or if he is directly forced to do so by his investors. The results make it
rather difficult to determine and recommend an optimal number of ratings that an investor should opt for when structuring a CDO. Therefore, it seems reasonable to address the rating outcomes of each rating agency in direct comparison with each other
in more detail.
Against the background of rating outcomes sorted by rating agencies, chapter 5 reveals that in the case of jointly rated CDO tranches, Fitch ratings are, on average, significantly lower (e.g. higher credit quality), as observed in the case of S&P and Moody’s ratings. Since Fitch is by far the smallest of the three rating agencies offering services in the field of CDO ratings, a potential explanation is seen in the form of selection bias: issuers only assign a CDO rating to Fitch if the expected outcome is better
than that obtained by Moody’s or S&P. In addition, it is also revealing to reflect on the
different rating methodologies. Fitch and S&P follow a PD based approach, whereas
Moody’s relies upon an EL based approach. In direct comparison, the rating outcomes
of Moody’s tend to reflect a lower credit quality, as documented both by Fitch and
S&P ratings. Thus, ratings based on an EL-based approach lead to lower ratings, as
observed in the case of PD-based rating processes.
Chapter 5 empirically explores the relevance of information agents with regard to the
pricing structure of CDS spreads. Consequently, it is argued - and also empirically
proven - that the level of information asymmetries - represented by the number of outstanding ratings - impacts CDO pricing structures. The findings show that, in addition
to other pricing factors (e.g. credit quality), as documented by the financial literature
130
Conclusion and Outlook
so far (see for example Longstaff and Rajan, 2008), the number of outstanding ratings
incorporates explanatory power with respect to the pricing structure of CDO credit
spreads. These results empirically support the argumentation stating that additional ratings reduce the existing information asymmetries between issuer and investor and thus
lower the credit spread premiums demanded by investors. A negative correlation between the number of outstanding ratings and credit spreads, in turn, allows us, first, to
detect the different levels of information asymmetries (e.g. single vs. double ratings)
and, second, reveals that higher levels of transparency - achieved through the participation of additional information agent(s) - eventually lowers the level of observed information asymmetries. Thus, it is noted that transparency can be viewed as an appropriated measure for increasing the trustworthiness of investors with regard to the secondary credit markets. Trustworthiness, in turn, should subsequently enhance market
functionality on a broader scale.
(III) Spill-over Effects between Stock Analysts’ Forecasts and CDS Spreads
Spill-over effects between stock analysts’ earnings forecasts and CDS spreads are analyzed with a strong focus on dynamic relationships throughout chapter 6. It is shown
(specifically for CDX-related entities) that higher stock analysts’ earnings forecasts are
associated with lower CDS spreads, whereas the dispersion of stock analysts’ earnings
forecasts are positively correlated with CDS spread levels. Thus, the role of stock analysts as information agents is not bound to the stock markets only but also affects CDS
spread levels. On the basis of a panel data analysis as well as a vector auto regression
model, no empirical evidence is found for any direction of existing lead-lag structures
taking place between the stock analysts’ forecasts and CDS spreads. However, significant lead-lag structures are detected between the dispersion of the stock analysts’ forecasts and CDS spreads, with the latter leading the first. With a positive correlation
structure existing between CDS spreads and the future dispersion of stock analysts’
forecasts, higher credit spread levels are followed by augmenting dispersion. Applying
a cointegration test, a long-run equilibrium relationship is additionally detected between the dispersion of stock analysts’ forecasts and CDS spreads. Additionally we
find significantly different patterns between U.S. and European entities. On average,
the spill-over effects between CDS spreads and dispersion of stock analysts’ forecasts
Conclusion and Outlook
131
for example seem to be more pronounced in the case of U.S. reference entities than for
European CDS contracts.
Chapter 6 sheds light on the issue of information asymmetries between two adjacent
capital markets and analyzes to which degree information asymmetries are bounded by
the borders of different sub segments of financial markets. A negative correlation
structure detected in the case of CDX-related reference entities between the stock analysts’ earnings forecasts and CDS spreads allows us to state that spill-over effects between these submarkets exist and that CDS spread levels and stock analysts’ forecasts
are associated with each other. As indicated beforehand, no lead-lag structures exist
between the stock analysts’ forecasts and CDS spreads. Thus, with regard to the
processing of new information of the underlying reference entities, co-movement takes
place on a non-lagged level. Since no specific adjustment processes (lead-lag structures) are detected, it is assumed that information is not misaligned between these two
subsets of capital markets but equally incorporated into CDS spread levels as well as
stock analysts’ forecasts. The results indicate that both CDS markets and stock analysts’ forecasts tend to rely on the same information sources and also seem to process
the content of the information simultaneously. Based on these findings it is additionally argued that none of the two has access to exclusive information. In addition, comovement between these adjacent market segments suggests that stock analysts function as information agents in the CDS markets and that their impact is thus not limited
to the stock markets only. Assuming a simple asset-based credit risk model (e.g. Merton, 1974), the direction of the detected correlation structures is also, from an economic perspective, rather reasonable: the decreasing future estimates lower the reference
entity’s value. A lower entity value correlates with higher credit risk levels of the underlying debt or higher CDS spreads respectively.
The dynamics between the dispersion of the stock analysts’ earnings forecasts and
CDS spreads, in turn, prove to be more revealing, since they are affected by lead-lag
structures. Co-movement is taking place on a lagged level with the corresponding
lead-lag dynamics regarded as information asymmetries: CDS spreads affect the future
formation of the dispersion of the stock analysts’ forecasts. CDS spreads, therefore,
also incorporate attributes relating to uncertainty among stock analysts, or – relying on
132
Conclusion and Outlook
the very basic perception of CDS spreads as a measure of risk – the dispersion of stock
analysts’ forecasts can be viewed against the background of the detected dynamics
with CDS spreads as a proxy for credit risk. The documented lead-lag relationship indicates that, in the case of the dispersion of stock analysts’ forecasts’ information incorporated into the CDS spreads impact - of course affected by a time lag – the formation of uncertainty among stock analysts (e.g. the dispersion of the stock analysts’
forecasts). Thus, the information distribution appears to be unequal regarding the dynamics between CDS spreads and the dispersion of the stock analysts’ forecasts.
Finally, chapter 6 also displays opposing patterns for European and U.S. reference entities. These findings can be interpreted as meaning that the CDS markets and stock
analysts follow different patterns with regard to regional affiliation. Since the U.S. follows a capital-market oriented regime, whereas the European financial markets are
aligned along a balance-sheet oriented market approach, it can be argued that the stock
analysts’ role is more pronounced in the U.S. than is observed for European reference
entities.
7.2 Relevance for Market Participants and Regulatory Authorities
Besides the dissertation’s contribution in the academic field, the empirical insights
gained throughout chapters 4 to 6 may also provide the market participants as well as
the regulatory authorities with valuable inside knowledge. The empirical evidence of
information asymmetries helps the regulatory authorities to track down problem areas.
Once the critical issues have been identified, the future regulatory standards can be defined in order to increase financial stability and revitalize activities in the secondary
credit markets.
The detection of rating model arbitrage emphasizes investors with regard to the specific structures of a CDO rating process as well as the different characteristics of the applied rating models. Investors learn that the selection of rating agencies throughout a
CDO transaction does not follow a random process but the rating agencies are chosen
with hindsight by the CDO issuers. In line with Fender and Kiff (2005), investors
should especially become curious if the tranches underlying the very same CDO trans-
Conclusion and Outlook
133
action are rated by the different rating agencies. In the course of chapter 4, the investor’s attention should be attired by the rather strong position of the issuer throughout
the CDO rating process. Since the investor (principal) maintains a direct relationship
with the issuer (agent), the investor should find himself in a position actually to challenge these market standards which promote information asymmetries throughout the
CDO rating process. Suggested changes initiated by the investors should therefore
specifically include the issuer’s commitment to fully publish its dialogue with the rating agencies. However, in order to successfully modify the market practices, investors
need to team up and combine the market power or persuade the regulatory authorities
to do so correspondingly.
From the issuer’s perspective, in turn, the current market standard of the CDO rating
process is rather agreeable, since he benefits from his comparably strong position.
However, if investors know about the potential impact of rating model arbitrage, they
might demand an additional risk premium in order to be compensated for the corresponding risk of rating model arbitrage. In this particular case, the issuer is even forced
to perform rating model arbitrage since he needs to compensate for the additional risk
premium demanded by the investor. Issuers should also cross-check whether it is indeed - as proposed by Fender and Kiff (2005) - reasonable to apply different rating
methodologies for different tranches of the CDO transaction. If so, there are additional
opportunities on the issuer’s side to enhance further its risk/return profile. The existing
CDO rating process was recently criticized by academics and politicians, as well as
market participants. In this context, the issuer-pays model as well as the consultancy
services offered by the rating agencies to the issuers were regarded as inappropriate
and even labeled as misleading.
Regulatory authorities seeking for potential areas of improvement can be attracted by
the phenomenon of rating model arbitrage and might use the empirical evidence as
outlined in chapter 4 as a starting point to apply new politics in order to increase
transparency in the secondary credit markets and thus reestablish trust among the market participants. Regulatory authorities could start a reasonable transparency initiative
by demanding as a first step that issuers disclose all of the communication they main-
134
Conclusion and Outlook
tain with the rating agencies relating to a specific CDO transaction. This would also
include dialogues with rating agencies not leading to a final rating publication.
The empirical evidence relating to the impact of multiple CDO ratings proves to be
quite revealing for the market participants. Investors learn that additional ratings lead
to lower credit spreads, which in turn can be interpreted as lower levels of information
asymmetries. Since the additional spread reduction is, on average, always above the
additional costs triggered by additional rating assignments, issuers benefit in either
way from multiple ratings and should, in principle, seek for at least two or even three
ratings. However, a detailed analysis of the level of each rating agency documented
that the Fitch ratings might be subject to a selection bias. It might not always be in the
investor’s interest actually to demand for example Fitch as a third rating. Investors
should, in any case, refrain from buying CDO tranches only rated by Fitch, since this
setting does expose them considerably to a potential selection bias. Moody’s rating
outcomes for the very same assets are in comparison to the ratings undertaken both by
Fitch and S&P linked to a lower credit quality. Thus, through a direct comparison, ratings relying on an EL based approach (Moody’s) are, from the investor’s perspective,
preferable to ratings accounting for a PD based approach (Fitch and S&P). In connection with empirical evidence of rating model arbitrage (see chapter 4), investors
should demand at least two ratings: one from each rating methodology. The issuers, in
turn, should try to persuade investors to accept a Fitch rating only. However, as the
applied data sample shows, single ratings by Fitch are rather seldom accepted by investors. Relying on the average notch differences observed in the case of jointly rated
CDO tranches, issuers should avoid issuance tranches with a rating of Moody only,
since Moody’s ratings tend to be the most conservative ones.
From the perspective of the regulatory authorities, the average notch differences are
also of particular interest. Even if chapter 5 provides us with empirical evidence that
notch differences are significant, the results also indicate that none of the rating agencies was as conservative as the recent months would have required. Throughout the
recent financial turmoil, all three rating agencies have been forced to downgrade the
ratings of their structured finance transactions on a broad scale. Both rating methodologies have therefore failed to foresee the dramatic downgrades and corresponding
Conclusion and Outlook
135
shortfalls as triggered by the so-called subprime crisis. None of the two rating methodologies turned out to be a superior model. Thus, the accreditation of one of the two
rating methodologies as the only benchmark cannot be regarded as an appropriate
strategy.
Against the background of the documented co-movement between the stock analysts’
forecasts and CDS spreads, the market participants being active in both sub segments
should acknowledge that changes in either dimension correlates with changes in the
other. Thus, protection seller and protection buyer should always factor in the fact that,
besides rating agencies also stock analysts can be viewed as additional information
agents, even if no lead lag structures are detected. Stock investors, in turn, should
watch closely the CDS spread levels. This is particularly true if stock investors want to
assess the benefit of future investments and the uncertainty among stock analysts. Diether et al. (2002) show that high levels of the dispersion of stock analysts’ forecasts
lead to higher abnormal returns. Thus, investors actually have an intelligible incentive
to estimate future levels of uncertainty among stock analysts regarding their forecasts.
With CDS spreads actually leading the dispersion of stock analysts’ forecasts, the CDS
markets might prove able to incorporate valuable guidance attributes in this context.
Finally, investors should be sensitized to substantially different patterns of comovement in the U.S. and Europe.
Since the issuer primarily benefits from existing information asymmetries on secondary credit markets, at a first glance he has no interest in increasing activity on the side
of the regulatory authorities to enhance transparency. However, the current financial
crisis has hit the secondary credit markets rather intensively. Thus, even the issuers
might therefore favour and ultimately benefit from an intervention by politics. Regulatory authorities in turn need to balance their intervention mechanisms in order not to
overdo their regulatory initiatives. As a comparable new sub segment, secondary credit
markets need to adjust their structures continuously. Fierce regulatory standards in turn
might harm this future evolution. However, given the current market conditions initiatives aiming at increased levels of transparency appear to be an appropriate response.
These initiatives do not need to be induced solely by regulatory authorities but could
also be co-launched by investors or issuers. In order to revitalize the secondary credit
136
Conclusion and Outlook
markets in the aftermath of the current financial turmoil, all three parties should therefore unite in order to lay the ground for future markets standards. Through increased
market standards and transparency applied improvements should take into account the
shortcomings of the past, but also reflect the high degree of dynamics associated with
market segments still at an early evolutionary stage. Since all three parties have a vital
interest in functioning secondary credit markets, all three should support initiatives in
order to increase market transparency and reduce information asymmetries subsequently.
References
XIX
References
Abdel-Khalik, A., Ajinkya, B. (1982): Returns to International Advantages: the Case
of Analysts’ Forecast Revisions, Accounting Review 57, 661-680.
Ackert, L. F., Athanassakos, G. (1997): Prior Uncertainty, Analyst Bias, and Subsequent Abnormal Returns, Journal of Financial Research 20, 263-273.
Ajinkya, B. B., Atiase, R. K., Gift, M. J. (1991): Volume of Trading and the Dispersion in Financial Analysts’ Earnings Forecasts, Accounting Review 66, 389-401.
Akerlof, G. A. (1970): The Market for “Lemons”: Quality Uncertainty and the Market
Mechanism, Quarterly Journal of Economics 84, 488-500.
Amir, E., Ganzach, Y. (1998): Overreaction and Underreaction in Analyst’s Forecasts, Journal of Economic and Organization 37, 333-347.
Arrow, K. J. (1969): The Organization of Economic Activity: Issues Pertinent to the
Choice of Market Versus Nonmarket Allocation, in: The Analysis and Evaluation of
Public Expenditure: The PPB System, 1, U.S. Joint Economic Committee, 91st Congress.1st Session, Washington, DC. 1969, 59-73.
Arrow, K. J. (1985): The Economics of Agency, in: Pratt, J. W., Zeckhauser, R. J.
(Ed.): Principals and Agents, Boston 1985, 37-51.
Ashkraft, A. B., Schuermann, T. (2008): Understanding the Securitization of Subprime Mortgage Credit, Staff Report No. 319, Federal Reserve Bank of New York,
March 2008.
Asquith, P., Mikhail, M. B., Au, A. S. (2005): Information Content of Equity Analyst
Reports, Journal of Financial Economics 75, 245-282.
Bader, H. (1996): Private Equity als Anlagekategorie: Theorie, Praxis und Portfoliomanagement für institutionelle Investoren, Diss., Bern o.P. 1996.
Baik, B., Park, C. (2003): Dispersion of Analysts’ Expectations and the Cross-section
of Stock Returns, Applied Financial Economics 13, 2003, 829-839.
XX
References
Balboa, M., Gómez-Sala, J. C., López-Espinosa, G. (2008): The Value of Adjusting
the Bias in Recommendations: International Evidence, European Financial Management forthcoming.
Bär, H. P. (1997): Asset Securitisation - Die Verbriefung von Finanzaktiven als innovative Finanzierungstechnik und neue Herausforderung für Banken, Diss., Bern o.P.
1997.
Basel Committee on Banking Supervision (2001): New Basel Capital Accord, Basel
Committee on Banking Supervision, 2001, www.bis.org.
Behrends, A. (1998): Informationsintermediation auf Finanzmärkten - Banken und
Rating-Agenturen als alternative Lösungsansätze, Diss., Bern o.P. 1998.
Beier-Middelschulte, A. (2004): Finanzkommunikation junger Emittenten - Reaktionen der Wirtschaftspresse auf Presse- und Ad hoc-Mitteilungen, Diss., Halle 2004.
Benkert, C. (2004): Explaining Credit Default Swap Premia, Journal of Futures Market 24, 71-92.
Bernet, B. (1982): Struktur und Prozess der betriebswirtschaftlichen Forschung, Diss.,
Zürich 1982.
Bernet, B. (2003): Institutionelle Grundlagen der Finanzintermediation, München
2003.
Bielecki, T. R., Rutkowski, M. (2002): Credit Risk: Modeling, Valuation and Hedging, Berlin o.P. 2002.
Billingsley, R., Lamy, R., Marr, M., Thompson, G. (1985): Split Ratings and Bond
Reoffering Yields, Financial Management Summer 1985, 59-65.
Blanco, R., Brennan, S., Marsh, I. W. (2005): An Empirical Analysis of the Dynamic Relation between Investment-Grade Bonds and Credit Default Swaps, Journal
of Finance 60, 2255-2281.
Bluhm, C., Overbeck, L. (2005): Risikoanalyse strukturierter Kreditprodukte, in:
Neupel, J., Rudolph, B., Hahnenstein, L. (Ed.): Aktuelle Entwicklungen im Bankencontrolling: Rating, Gesamtbanksteuerung und Basel II, Düsseldorf 2005, 119-144.
References
XXI
Bolliger, G. (2004): The Characteristics of Individual Analyst’s Forecasts in Europe,
Journal of Banking and Finance 28, 2283-2309.
Borio, C. (2008): The Financial Turmoil of 2007?: A Preliminary Assessment and
Some Policy Considerations, BIS Working Paper No. 251, Bank for International
Settlements.
Bradshaw, M. T., Richardson, S. A., Sloan, R. G. (2006): The Relation between
Corporate Financing Activities, Analysts’ Forecasts and Stock Returns, Journal of
Accounting and Economics 42, 53-85.
Breuer, W. (1995): Finanzintermediation und Reputationseffekte, Kredit und Kapital
28, 516-533.
Brown, L. d. (2001): How Important Is Past Analyst Forecast Accuracy?, Financial
Analysts Journal 57, 44-49.
Buiter, W.H. (2007): Lessons from the 2007 Financial Crisis, Policy Insight No. 18,
Centre for Economic Policy Research, 1-17.
Cantor, R., Packer, F. (1995): The Credit Rating Industry, Journal of Fixed Income
5, 10-34.
Cantor, R., Packer, F. (1997): Differences of Opinion and Selection Bias in the Credit Rating Industry, Journal of Banking and Finance 21, 1395-1417.
Capstaff, J., Paudyal, K., Rees, W. (2001): A Comparative Analysis of Earnings
Forecasts in Europe, Journal of Business Finance and Accounting 28, 531-562.
Carter, J., Osako, C., Cunnigham, T., Hardee, R. (2006): CDO Structures and Definitions, Fitch Ratings, London 2006.
CESR (2008a): The Role of Rating Agencies in Structured Finance, Consultation Paper, Committee of European Securities Regulators, February 2008.
CESR (2008b): CESR’s Second Report to the European Commission: On the Compliance of Credit Rating Agencies with the IOSCO code and the Role of Rating
Agencies in Structured Finance, Consultation Paper, Committee of European Securities Regulators, May 2008.
XXII
References
Chacko, G., Sjoman, A., Motohashi, H., Dessain, V.: Credit Derivatives - A Primer
on Credit Risk, Modeling, and Instruments, Upper Saddle River 2006.
Chae, J. (2005): Trading Volume, Information Asymmetry, and Timing Information,
Journal of Finance, 60, 413-442.
Chan, L. K. C., Karceski, J., Lakonishok, J. (2007): Analysts’ Conflicts of Interest
and Biases in Earnings Forecasts, Journal of Financial and Quantitative Analysis 42,
893-914.
Chan-Lau, J. A., Lu, Y. (2006): Idiosyncratic and Systematic Risk in the European
Corporate Sector: A CDO Perspective, Working Paper 06/107, International Monetary Fund.
Chan-Lau, J., Ong, L. (2006): The Credit Risk Transfer Market and Stability Implications for U.K. Financial Institutions, Working Paper, International Monetary Fund.
Chirico, A. (2008): Credit Rating Agencies: The Main Lacuna in EU Regulation Governing Conflicts of Interest, ECMI Commentary No. 17, Centre for European Policy
Studies.
Clancy, P. (2006): Legal Structure of Cash and Synthetic Transactions, in: Watson,
R., Carter, J. (Ed.): Asset Securitisation and Synthetic Structures - Innovations in the
European Credit Markets, London 2006, 37-50.
Coase, R. (1937): The Nature of the Firm, Economica 4, 386-405.
Cowen, A., Groysberg, B., Healy, P. (2006): Which Types of Analyst Firms Are
More Optimistic?, Journal of Accounting and Economics 41, 119-146.
Cox, C. (2008): Turmoil in U.S. Credit Markets. Recent Actions Regarding Government Sponsored Entities, Investment Banks and Other Financial Institutions, Testimony before the Committee on Banking, Housing, and Urban Affairs United States
Senate, 23. September 2008.
Daniels, K. N., Jensen, M. S. (2005): The Effect of Credit Ratings on Credit Default
Swap Spreads and Credit Spreads, Journal of Fixed Income 15, 16-33.
Diamond, D. W. (1984): Financial Intermediation and Delegated Monitoring, Review
of Economic Studies 51, 393-414.
References
XXIII
Dichev, I. D., Piotroski, J. D. (2001): The Long-Run Stock Returns Following Bond
Ratings Changes, Journal of Finance 56, 173-203.
Diether, K. B., Malloy, C. J., Scherbina, A. (2002): Differences of Opinion and the
Cross Section of Stock Returns, Journal of Finance 57, 2113-2141.
Dische, A. (2002): Dispersion in Analyst Forecasts and the Profitability of Earnings
Momentum Strategies, European Financial Management 8, 211-228.
Doukas, J. A., Kim, C., Pantzalies, C. (2006): Divergence of Opinion and Equity Returns, Journal of Financial and Quantitative Analysis 41, 573-606.
Duffie, D., Malamud, S., Manso, G. (2009): The Relative Contributions of Private
Information Sharing and Public Information Releases to Information Aggregation,
Working Paper, Graduate School of Business, Stanford University.
Duhon, T. (2006): The Market For First Loss And Mezzanine Risk, in: Watson, R.,
Carter, J. (Ed.): Asset Securitisation and Synthetic Structures – Innovations in the
European Credit Markets, London 2006, 126-138.
Ederington, L. H., Goh, J. C. (1998): Bond Rating Agencies and Stock Analysts;
Who knows What When?, Journal of Financial and Quantitative Analysis 33, 569585.
Effenberger, D. (2004): Credit Derivatives: Effects on the Stability of Financial Markets, Deutsche Bank Research, June 2004.
Engsted, T. and Tanggard, C. (2004): The comovement of US and UK stocks, European Financial Management 10, 593-607.
Erlei, M., Jost, P. (2001): Theoretische Grundlagen des Transaktionskostenansatzes,
in: Jost, P. (Ed.): Der Transaktionskostenansatz in der Betriebswirtschaftslehre,
Stuttgart 2001, 35-75.
ESME (2008): Role of Credit Rating Agencies, Report to the European Commission,
European Securities Markets Expert Group, June 2008.
EU Commission (2008): Proposal for a Regulation of the European Parliament and of
the Council on Credit Rating Agencies, Commission of the European Communities,
12. November 2008.
XXIV
References
Fama, E. F. (1980): Agency Problems and the Theory of the Firm, Journal of Political
Economy 88, 288-307.
Fama, E. F. (1991): Efficient Markets II, in: Journal of Finance, Vol. 46, 1991, S.
1575-1617
Fama, E. F., Jensen, M. C. (1983): Separation of Ownership and Control, Journal of
Law and Economics 26, 301-325.
Fazzari, S. M., Athey, M. (1987): Asymmetric Information, Financing Constraints,
and Investment, Review of Economics and Statistics 69, 481-487.
Fender, I., Kiff, J. (2005): CDO Rating Methodology: Some Thoughts on Model Risk
and Its Implications, Journal of Credit Risk 1, 37-58.
Fitch Derivatives (2006): Global Rating Criteria for CDOs, Fitch Ratings.
Flannery, M. (1986): Asymmetric Information and Risky Debt Maturity Choice,
Journal of Finance 49, 19-37.
Ford, R. (2006): Exploring the Investor Base: Who Buys and Sells European ABS
and Why?, in: Watson, R., Carter, J. (Ed.): Asset Securitisation and Synthetic Structures - Innovations in the European Credit Markets, London 2006, 99-125.
Franke, G., Krahnen, J. P. (2008): The Future of Securitization, Working Paper,
University of Konstanz.
Freixas, X., Rochet, J. (1997): Microeconomics of Banking, London 1997.
Fried, D., Givoly, D. (1982): Financial Analysts’ Forecasts of Earnings - A Better
Surrogate for Market Expectations, Journal of Accounting and Economics 4, 85-107.
Frost, C. A. (2006): Credit Rating Agencies in Capital Markets: A Review of Research Evidence on Selected Criticisms of the Agencies, Working Paper, State University of New York, March 2006.
Gebhardt, W. R. ,Lee, C. M. C., Swaminathan, B. (2001): Toward an implied cost
of capital, Journal of Accounting Research 39, 135-176.
Gerster, K. (2005): Informationsasymmetrien im Finanzdienstleistungsbereich unter
spezieller Betrachtung von Alternative Investments, Diss., Bern o.P. 2005.
References
XXV
Goh, J. C., Ederington, L. H. (1993): Is a Bond Rating Downgrade Bad News, Good
News, or No News for Stockholders?, Journal of Finance 48, 2001-2008.
Goodman, L. S., Fabozzi, F. J. (2002): Collateralized Debt Obligations: Structures
and Analysis, Hoboken 2002.
Gorton, G. B., Pennacchi, G. G. (1995): Banks and Loans Sales - Marketing Nonmarketable Assets, Journal of Monetary Economics 35, 389-411.
Gorton, G., Soleles, N. S. (2005): Special Purpose Vehicles and Securitization, Working Paper 11190, National Bureau of Economic Research, March 2005.
Goswami, G., Noe, T., Rebello, M. (1995): Debt Financing under Asymmetric Information, Journal of Finance 50, 633-659.
Gregory, J. (2004): Structured Credit Derivatives and Portfolio Management - Introduction, in: Gregory, J. (Ed.): Credit Derivatives: The Definitive Guide, London
2004, 151-156.
Güttler, A. (2008): Lead-lag Relationships and Rating Convergence Among Credit
Rating Agencies, Working Paper, European Business School, International University Schloss Reichartshausen.
Hager, S. (2008): Pricing Portfolio Credit Derivatives by Means of Evolutionary Algorithms, Diss., Tübingen 2008.
Hand, J. R. M., Holthausen, R. W., Leftwich, R. W. (1992): The Effect of Bond
Rating Agency Announcements on Bond and Stock Prices, Journal of Finance 47,
733-752.
Hartmann-Wendels, T., Pfingsten, A., Weber, M. (2007): Bankbetriebslehre, 4th
Edition, Berlin o.P., 2007.
Heinke, V. G. (1998): Bonitätsrisiko und Credit Rating festverzinslicher Wertpapiere,
Diss., Münster 1998.
Holthausen, R. W., Leftwich, R. W. (1986): The Effect of Bond Rating Changes on
Common Stock Prices, Journal of Financial Economics 17, 57-89.
XXVI
References
Hsueh, L., Kidwell, S. (1988): Bond Ratings: Are Two Better Than One? Financial
Management Spring 1988, 46-53.
Hu, J. (2008): 2008 U.S. CDO Outlook and 2007 Review: Issuance Down in 2007
Triggered by Subprime Mortgages Meltdown; Lower Overall Issuance Expected in
2008, Moody’s Investors Service, March 2008.
Hu, J. and Cantor, R. (2006): The Relationship between Issuance Spreads and Credit
Performance of Structured Finance Securities, Journal of Fixed Income 16, 5-20.
Hull (2006): Options, Futures, and Other Derivatives, Sixth Edition, New Jersey 2006.
Hull, J., Predescu, M., White, A. (2004): The Relationship between Credit Default
Swap Spreads, Bond Yields, and Credit Rating Announcements, Journal of Banking
and Finance 28, 2789-2811.
Hull, J., White, A. (2004): Valuation of a CDO and an n-th Default CDS without
Monte Carlo Simulation, Journal of Derivatives Winter 2004, 8-23.
IOSCO (2008a): The Role of Credit Rating Agencies in Structured Finance Markets,
Final Report, Technical Committee of the International Organization of Securities
Commissions, May 2008.
Jensen, M. C., Meckling, W. H. (1976): Theory of the Firm: Managerial Behavior,
Agency Costs and Ownerships Structure, Journal of Financial Economics 3, 305-360.
Jewell, J., Livingston, M. (1998): Split Ratings, Bond Yields, and Underwriter
Spreads, Journal of Financial Research 21, 185-204.
Jewell, J., Livingston, M. (1999): A Comparison of Bond Ratings from Moody’s,
S&P and Fitch IBCA, Financial Markets, Institutions and Instruments 8, 1-45.
Jobst, A. A. (2006): European Securitization: A GARCH Model of Secondary Market
Spreads, Journal of Structured Finance 12, 55-80.
Johansen, S. (1991): Estimation and Hypothesis Testing of Cointegration Vectors in
Gaussian Vector Autoregressive Models, Econometrica 59, 1551–1580.
Johansen, S. (1995): Likelihood-based Inference in Cointegrated Vector Autoregressive Models, Oxford University Press, Oxford.
References
XXVII
Johnson, T. C. (2004): Forecast Dispersion and the Cross Section of Expected Returns, Journal of Finance 59, 1957-1978.
Jorion, P. (2007): Value at Risk, 3rd Edition, New York o.P., 2007.
Jortzik, S. (2005): Semi-analytische und simulative Kreditrisikomessung synthetischer Collateralized Debt Obligations bei heterogenen Referenzportfolios, Diss., Göttingen 2005.
Katz, S. (1974): The Price Adjustment Process of Bonds to Rating Reclassifications:
A Test of Bond Market Efficiency, Journal of Finance 29, 551-559.
Kazemi, H. B. (1991): Dispersion of Beliefs, Asset Prices, and Noisy Aggregation of
Information, Financial Review 26, 1-13.
Kiener, S. (1990): Die Principal-Agent-Theorie aus informationsökonomischer Sicht,
Diss., Heidelberg 1990.
Kish, R., Hogan, K., Olson, G. (1999): Does the market perceive a difference in rating agencies?, Quarterly Review of Economics and Finance 39, 363-377.
Kliger, D., Sarig, O. (2000): The Information Value of Bonds Ratings, Journal of
Finance 55, 2879-2902.
Krinsman, A.N. (2007): Subprime Mortgage Meltdown: How Did It Happen and
How Will It End?, Journal of Structured Finance 13, 13-19.
L’Her, J., Suret, J. (1996): Consensus, Dispersion and Security Prices, Contemporary
Accounting Research 13, 209-228.
Larosière Report (2009): The High Level Group on Financial Supervision in the EU,
chaired by Jacques de Larosière, Brussels, 25.02.2009.
Lehnert, T., Neske, f. (2006): On the Relationship Between Credit Rating announcements and Credit Default Swap Spreads for European Reference Entities, Journal of
Credit Risk 2, 83-90.
Leland, H. E., Pyle, D. H. (1977): Informational Asymmetries, Financial Structure,
and Financial Intermediation, Journal of Finance 32, 371-387.
XXVIII
References
Lindtner, A. (2006): Asset Backed Securities - Ein Cash flow-Model, 2. Edition,
Sternenfels 2006.
Liu, P., Moore, T. (1987): The Impact of Split Bond Ratings on Risk Premia, Financial Review 22, 71.85.
Longstaff, F. A., Rajan, A. (2008): An Empirical Analysis of the Pricing of Collateralized Debt Obligations, Journal of Finance 63, 529-563.
Longstaff, F.A., Mithal, S., Neis, E. (2003): The credit-default swap market: is credit
protection priced correctly?, Working Paper, University of California, 2003.
Lucas, D. J., Goodman, L. S., Fabozzi, F. J. (2006): Collateralized Debt Obligations
- Structures and Analysis, Second Edition, Hoboken 2006.
Lucas, D., Goodman, L. S., Fabozzi, F. J., Manning, R. J. (2007): Developments in
Collateralized Debt Obligations - New Products and Insights, New Jersey 2007.
Lucas, D.J., Goodman, L.S., Fabozzi, F.J. (2008): How to Save the Rating Agencies, Journal of Structured Finance 14, 21-26.
Lys, T., Sohn, S. (1990): The Association Between Revisions of Financial Analysts’
Earnings Forecasts and Security-Price Changes, Journal of Accounting and Economics 13, 341-363.
MacKinnon, J. G., Haug, A. A., Michelis, L. (1999): Numerical Distribution Functions of Likelihood Ratio Tests For Cointegration, Journal of Applied Econometrics
14, 563-577.
Manns, J. (2008): Rating Risk after the Subprime Mortgage Crisis: A User Fee Approach for Rating Agency Accountability, Working Paper, North Carolina Law Review Forthcoming.
Martiensen, J. (2000): Institutionenökonomik, München 2000.
Martin, M. R. W., Reitz, S., Wenn, C. S. (2006): Kreditderivate und Kreditrisikomodelle, Wiesbaden 2006.
References
XXIX
Mason, J. R., Rosner, J. (2007): Where Did the Risk Go? How Misapplied Bond Ratings cause Mortgage Backed Securities and Collateralized Debt Obligation Market
Disruption, Working Paper, Drexel University, May 2007.
Merton, R. C. (1974): On the Pricing of Corporate Debt: The Pricing Structure of Interest Rates, Journal of Finance 29, 449-470.
Metz, A. (2007): On the Prediction of Credit Ratings, Moody’s Investors Service,
March 2007.
Milleker, D., Sauerschell, S. (2008): Gebrauchtwagen à la Wall Street, Financial Times Deutschland, 20. August 2008, 26.
Miller, E. M. (1977): Risk, Uncertainty, and Divergence of Opinion, Journal of
Finance 32, 1151-1168.
Mirrless, J. A. (1999): The Theory of Moral Hazard and Unobservable Behaviour:
Part I, Review of Economic Studies 66, 3-21.
Moody’s (2007a): Comparing Ratings on Jointly-Rated U.S. Structured Finance Securities: 2007 Update; Moody’s Structured Finance Special Report 2007.
Moody’s (2007b): Comment Letter on Proposed Rules regarding Oversight of Credit
Rating Agencies registered as Nationally Recognized Statistical Rating Organizations, Securities and Exchange Commission.
Morkötter, S., Westerfeld, S. (2007): Ratingagenturen haben ein Kommunikationsproblem, Kredit & Rating Praxis, 6/2007, 9-12.
Morkötter S., Westerfeld, S. (2008): Asset Securitisation - Die Geschäftsmodelle
von Ratingagenturen im Spannungsfeld einer Principal-Agent-Betrachtung, Zeitschrift für das gesamte Kreditwesen 61, 393-396.
Morkötter, S., Westerfeld, S. (2009a): Impact of Multiple CDO Ratings on Credit
Spreads, Journal of Fixed Income 19, 55-75.
Morkötter, S., Westerfeld, S. (2009b): Rating Model Arbitrage in CDO Markets: An
Empirical Analysis, International Review of Financial Analysis 18, 21-33.
XXX
References
Morr, M., Schneider, N., Bräutigam A., Hippeli, P. (2005): Zukünftige strategische
Optionen im Kreditgeschäft am Beispiel von Asset-Backed Securities, Aachen 2005.
Morse, D., Stephan, J., Stice, E. K. (1991): Earnings Announcements and the Convergence (or Divergence) of Beliefs, Accounting Review 66, 376-388.
Müller-Stewens, G., Lechner, C. (2003): Strategisches Management - Wie strategische Inititativen zum Wandel führen, Second Edition, Stuttgart 2003.
Norden, L. (2004): Kreditderivate: Zwischen Kapitalmarkt und bankbetrieblicher
Verwendung, Diss., Mannheim 2004.
Norden, L., Weber, M. (2004): Informational Efficiency of Credit Default Swaps and
Stock Markets: The Impact of Credit Rating Announcements, Journal of Banking and
Finance 28, 2813-2843.
Norden, L., Weber, M. (2004): Informational Efficiency of Credit Default Swaps and
Stock Markets: The Impact of Credit Rating Announcements, Journal of Banking and
Finance 28, 2813-2843.
O’Kane, D. (2001): Credit Derivatives Explained: Market, Products and Regulations,
Structured Credit Research, Lehman Brothers, New York 2001.
Oster, N. (2007): Biased Reactions to Information and Learning in Analysts’ Forecasts, Working Paper, Stanford University, December 2007.
Park, C. (2005): Stock Return Predictability and the Dispersion in Earnings Forecasts,
Journal of Business 78, 2351-2375.
Partnoy, F. (2006): How and Why Credit Rating Agencies are Not Like other Gatekeepers, Research Paper No. 07-46, University of San Diego, May 2006.
Patz, D. H. (1989): UK Analysts’ Earnings Forecasts, Accounting and Business Research 19, 267-275.
Pawley, M. (2004): Collateralised Debt Obligations - Balance Sheet and Arbitrage
CDOs, Deutsche Bank.
References
XXXI
Peretyatkin, V., Perraudin, W. (2002): EL and DP Approaches to Rating CDOs and
the Scope for Ratings Shopping, in: Ong, M. K. (Eds.), Credit Ratings – Methodologies, Rationale and Default Risk, London 2002, 495-506.
Perraudin, W., Taylor, A. P. (2004): On the Consistency of Ratings and Bond Market Yields, Journal of Banking and Finance 28, 2769-2788.
Perry, L., Liu, P., Evans, D. (1988): Modified Bond Ratings: Further Evidence on
the Effect of Split Ratings on Corporate Bond Yields, Journal of Business Finance
and Accounting 15, 231-241.
Peterson, D., Peterson, P. (1982): The Effect of Changing Expectations Upon Stock
Returns, Journal of Financial and Quantitative Analysis 17, 799-813.
Picot, A., Maier, M. (1993): Information als Wettbewerbsfaktor, in: Pressmar, D. B.
(Ed.): Informationsmanagement, Wiesbaden 1993, 31-53.
Rajendra, R. G., O’Toole, C., Pahlson-Möller, I., Pratt, J. (2008): European Asset
Backed Barometer - Weekly Market Monitor, Global Markets Research, Deutsche
Bank, November 2008.
Ramakrishan, R. T. S., Thakor, A. V. (1984): Information Reliability and a Theory
of Financial Intermediation, Review of Economic Studies 51, 415-432.
Reiter, S., Ziebart, D. (1991): Bond Yields, Ratings, and Financial Information: Evidence from Public Utility Issues, Financial Review 26, 45-73.
Richardson, S., Hong Teoh, S., Wysocki, P. D. (2004): The Walk-down to Beatable
Analyst Forecasts: The Role of Equity Issuance and Insider Trading Incentives, Contemporary Accounting Research 21, 885-924.
S&P (2007): Das Geschäftsmodell einer internationalen Ratingagentur, Presentation at
the University of St. Gallen by Thorsten Hinrichs, 4 September 2007.
Schiefer, D. (2008): Collateralized Debt Obligations (CDOs) - Eine empirische Analyse der Bonitätsrisikoprämie auf Finanzmärkten, Diss., Bad Sodem/ Ts. 2008.
SEC (2008): Summary Report of Issues Identified in the Commission Staff’s Examinations of Select Credit Rating Agencies, United States Securities and Exchange
Commission, July 2008.
XXXII
References
Signer, A. (2003): Generieren Hedge Funds einen Mehrwert?, Diss., Bern o.P. 2003.
Sinclair, P., Spier, G. , Skinner, T. (2008): Bonuses, Credit Rating Agencies and the
Credit Crunch, Conference Paper 2008, Centre for Dynamic Macroeconomic Analysis, University of St. Andrews, September 2008.
Spremann, K. (1990): Asymmetrische Information, Zeitschrift für Betriebswirtschaft
60, 561-586.
Stickel, S. E. (1991): Common Stock Returns Surrounding Earnings Forecast Revisions: More Puzzling Evidence, Accounting Review 66, 402-416.
Stickney, C. P., Brown, P. R., Wahlen, J. M. (2007): Financial Reporting, Financial
Statement Analysis, and Valuation - A Strategic Perspective, Sixth Edition, Mason
2007.
Sufi, A. (2007): Information Asymmetry and Financing Arrangements: Evidence from
Syndicated Loans, Journal of Finance 62, 629-668.
Tavakoli, J. M. (2003): Collateralized Debt Obligations and Structured Finance New Developments in Cash and Synthetic Securitization, Hoboken 2003.
Taylor, F. (2007): Mastering Derivatives Markets - A Step-by-Step Guide to the
Products, Applications and Risks, Third Edition, Harlow 2007.
Trezzini, L. (2005): Finanzierungsstrukturierung im Venture Capital, Diss., Bern o.P.
2005.
Umlauf, S. R. (1991): Information Asymmetries and Security Market Design: An
Empirical Study of the Secondary Markets for U.S. Government Securities, Journal
of Finance 46, 929-953.
Vink, D., Thibeault, A. E. (2008): ABS, MBS and CDO Compared: An Empirical
Analysis, Journal of Structured Finance 14, 27-45.
von Thadden, E. (1995): Long-Term Contracts, Short-Term Investment and Monitoring, Review of Economic Studies 62, 557-575.
Warburg (2006): Equity Research Report: BASF, M.M. Warburg & CO, 10 August
2006.
References
XXXIII
Watson, R., Carter, J. (2006): Asset Securitisation and Synthetic Structures - Innovations in the Credit Markets, London 2006.
Weinstein, M. I. (1977): The Effect of a Rating Change Announcement on Bond
Price, Journal of Financial Economics, 329-350.
Westerfeld, S. (2004): Kreditportfoliomanagement im Wandel, Diss., Bern o.P. 2004.
Westerfeld, S. (2008): Financial Services Management - Lehreinheit 7 - Asset Backed
Finance, St. Gallen 2008.
Williamson, O. E. (1975): Markets and Hierarchies: Analysis and Antitrust Implications, New York 1975.
Williamson, O. E. (1985): The Economic Institutions of Capitalism, New York 1985.
Zuberbühler, D. (2008): Globalisierte Kreditkrise - Konsequenzen für die Bankenaufsicht, Jahresmedienkonferenz, April 2008, Bern.
Internet Sources:
Cohen (2008): Independent Research Analyst Coverage, Cohen Independent Research
Group, www.cohenresearch.com, 27.11.2008.
Deutsche Bank (2008): Interview with Dr. Josef Ackermann, Deutsche Bank,
www.db.com/de/ content/company/nachrichten_9899.htm, 12.11.2008.
Fitch (2009): Fitch Ratings, www.fitchratings.com, 23.03.2009.
ISDA (2008): ISDA Market Survey Historical Data 1987-2008, International Swaps
and Derivative Association, www.isda.org, 18.11.2008.
LSTA (2008): Loan Syndications and Trading Association, www.lsta.org, 27.11.2008.
Moody’s (2009): Moody’s Investors Services, www.moodys.com, 23.03.2009.
S&P (2008): CDO Spotlight: Major Global Cash Flow CDO Managers Now Includes
Both U.S. And European Managers, Standard & Poor’s http://www2. standardandpoors.com/ portal/site/sp/en/us/page.article/3,1,1,0,1148450915608.html, 18.11.2008.
XXXIV
References
S&P (2009): Standard & Poor’s CDO Interface, www.sp.cdointerface.com,
23.03.2009
Appendix
XXXV
Appendix
Appendix I: Reference Entities
No.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
Reference Entity
ABB International Finance Limited
Accor
ACE Limited
Adecco S.A.
Aegon N.V.
Aetna
Aktiebolaget Volvo
AKZO Nobel N.V.
Alcatel Lucent
Alcoa
Allianz SE
Altria Group
Amerada Hess Corporation
American Electronic Power Company
American Express Company
American International Group
Anadarko Petroleum Corporation
Anglo American Plc
Arrow Electronics
ArvinMeritor
Assicurazioni Generali - Societa per Azioni
AutoZone
BAE Systems
Banca Monte dei paschi di siena S.P.A.
Banco Bilbao Vizcaya Argentaria, Sociedad Anonima
Banco Comercial Portugues S.A.
Banco Espirito Santo S.A.
Banco Santander S.A.
BASF SE
Baxter International
Bayer Aktiengesellschaft
BNP Paribas
Boeing Capital Corporation
BP P.L.C.
Bristol-Myers Squibb Company
British Airways Inc.
Burlington Northern Santa Fe Corporation
Cable and Wireless Public Limited Company
Campbell South Company
Capital One Bank
Cardinal Health
Carnival Corporation
Carrefour
Casino Guichard-Perrachon
Caterpillar Inc.
Centex Corporation
CenturyTel
Comcast Corporation
Commerzbank
COMPAGNIE DE SAINT-GOBAIN
Computer Associates International
Computer Sciences Corporation
ConAgra Foods
ConocoPhillips Corporation
Constellation Energy Group
Continental Aktiengesellschaft
Cooper Tire & Rubber Company
Credit Agricole
Credit Suisse Group
^yŽƌƉŽƌĂƚŝŽŶ
Cummins Inc.
CVS Caremark Corporation
Daimler AG
Deere & Company
Deutsche Bank
Deutsche Telekom
Devon Energy Corporation
Country
Switzerland
France
United States
Switzerland
Netherlands
United States
Sweden
Netherlands
France
United States
Germany
United States
United States
United States
United States
United States
United States
UK
United States
United States
Italy
United States
UK
Italy
Spain
Portugal
Portugal
Spain
Germany
United States
Germany
France
United States
UK
United States
UK
United States
UK
United States
United States
United States
United States
France
France
United States
United States
United States
United States
Germany
France
United States
United States
United States
United States
United States
Germany
United States
France
Switzerland
United States
United States
United States
Germany
United States
Germany
Germany
United States
Sector
Industrials
TMT
Financials
Consumer
Financials
Financials
Industrials
Industrials
TMT
Industrials
Financials
Consumer
Industrials
Energy
Financials
Financials
Energy
Industrials
TMT
Industrials
Financials
Consumer
Industrials
Financials
Financials
Financials
Financials
Financials
Industrials
Consumer
Consumer
Financials
Industrials
Energy
Consumer
Consumer
Industrials
TMT
Consumer
Financials
Consumer
Consumer
Consumer
Consumer
Industrials
Consumer
TMT
TMT
Financials
Industrials
TMT
TMT
Consumer
Energy
Energy
Industrials
Industrials
Financials
Financials
Industrials
Industrials
Consumer
Industrials
Industrials
Financials
TMT
Energy
Index Group
iTraxx
iTraxx
y
iTraxx
iTraxx
y
iTraxx
iTraxx
iTraxx
y
iTraxx
y
y
y
y
y
y
iTraxx
y
y
iTraxx
y
iTraxx
iTraxx
iTraxx
iTraxx
iTraxx
iTraxx
iTraxx
y
iTraxx
iTraxx
y
iTraxx
y
iTraxx
y
iTraxx
y
y
y
y
iTraxx
iTraxx
y
y
y
y
iTraxx
iTraxx
y
y
y
y
y
iTraxx
y
iTraxx
iTraxx
y
y
y
iTraxx
y
iTraxx
iTraxx
y
XXXVI
Appendix
Appendix I - Continued
No.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
82.
83.
84.
85.
86.
87.
88.
89.
90.
91.
92.
93.
94.
95.
96.
97.
98.
99.
100.
101.
102.
103.
104.
105.
106.
107.
108.
109.
110.
111.
112.
113.
114.
115.
116.
117.
118.
119.
120.
121.
122.
123.
124.
125.
126.
127.
128.
129.
130.
131.
132.
133.
134.
135.
136.
137.
138.
Reference Entity
Dominion Resources
DSG International Plc
Duke Energy Carolinas
E.I. du Pont de Nemours and Company
E.ON AG
Eastman Chemical Company
Eastman Kodak Company
EDP - Energias de Portugal S.A.
Enel S.P.A.
FIAT S.P.A.
FINMECCANICA S.P.A.
FirstEnergy Corporation
Ford Motor Credit Company
France Telecom
Gannett Co. Inc.
General Electric Capital Corporation
General Mills
General Motors Acceptance Corporation
Goodrich Corporation
Goodyear Tire & Rubber Company
Groupe Danone
Halliburton Company
Hannover Rueckversicherung AG
HeidelbergCement AG
Hellenic Telecommunications Organisation Societe Anonyme
Henkel AG & Co. KGaA
Hewlett-Packard Company
Honeywell International
HSBC Bank Plc
Iberdrola
International Business Machines Corporation
International Paper Company
INTESA SANPAOLA SPA
Jones Apparel Group
Kingfisher
Koninklijke Ahold N.V.
Koninklijke DSM N.V.
Koninklijke KPN N.V.
Koninklijke Philips Electronics N.V.
Kraft Foods Inc.
Ladbrokes Plc
LAFARGE
LEAR Corporation
Lennar Corporation
Limited Brands
Linde
Lloyds TSB Bank Plc
Lockheed Martin Corporation
Lufthansa
LVMH Moet Hennessy Louis Vuitton
Macy´s Inc.
Marriott International Inc.
Masco Corporation
McDonald´s Corporation
MeadWestvaco Corporation
MetLife
MGM Mirage
Motorola
Muenchener Rueckversicherungs-Gesellschaft AG
Nestle S.A.
Newell Rubbermaid
Nordstorm
Norfolk Southern Corporation
Northrop Grumman Corporation
Olin Corporation
Omnicom Group
Peugeot SA
Portugal Telecom International Finance B.V.
PPR
Progress Energy
Pulte Homes
Country
United States
UK
United States
United States
Germany
United States
United States
Portugal
Italy
Italy
Italy
United States
United States
France
United States
United States
United States
United States
United States
United States
France
United States
Germany
Germany
Greece
Germany
United States
United States
UK
Spain
United States
United States
Italy
United States
UK
Netherlands
Netherlands
Netherlands
Netherlands
United States
UK
France
United States
United States
United States
Germany
UK
United States
Germany
France
United States
United States
United States
United States
United States
United States
United States
United States
Germany
Switzerland
United States
United States
United States
United States
United States
United States
France
Portugal
France
United States
United States
Sector
Energy
Consumer
Energy
Industrials
Energy
Industrials
Industrials
Energy
Energy
Industrials
Industrials
Energy
Autos/ Financials
TMT
TMT
Industrials
Industrials
Autos/ Financials
Industrials
Industrials
Consumer
Energy
Financials
Industrials
TMT
Consumer
TMT
Industrials
Financials
Energy
TMT
Industrials
Financials
Consumer
Consumer
Consumer
Industrials
TMT
Consumer
Consumer
Consumer
Industrials
Industrials
Industrials
Consumer
Industrials
Financials
Industrials
Consumer
Consumer
Consumer
Consumer
Industrials
Consumer
Industrials
Financials
Consumer
TMT
Financials
Consumer
Consumer
Consumer
Industrials
Industrials
Industrials
TMT
Industrials
TMT
Consumer
Energy
Consumer
Index Group
y
iTraxx
y
y
iTraxx
y
y
iTraxx
iTraxx
iTraxx
iTraxx
y
y
iTraxx
y
y
y
y
y
y
iTraxx
y
iTraxx
iTraxx
iTraxx
iTraxx
y
y
iTraxx
iTraxx
y
y
iTraxx
y
iTraxx
iTraxx
iTraxx
iTraxx
iTraxx
y
iTraxx
iTraxx
y
y
y
iTraxx
iTraxx
y
iTraxx
iTraxx
y
y
y
y
y
y
y
y
iTraxx
iTraxx
y
y
y
y
y
y
iTraxx
iTraxx
iTraxx
y
y
Appendix
XXXVII
Appendix I - Continued
No.
139.
140.
141.
142.
143.
144.
145.
146.
147.
148.
149.
150.
151.
152.
153.
154.
155.
156.
157.
158.
159.
160.
161.
162.
163.
164.
165.
166.
167.
168.
169.
170.
171.
172.
173.
174.
175.
176.
177.
178.
179.
180.
181.
182.
183.
184.
185.
186.
187.
188.
189.
190.
191.
192.
193.
194.
195.
196.
197.
198.
199.
200.
201.
202.
203.
204.
Reference Entity
RadioShack Corporation
Raytheon Company
Repsol YPF S.A.
Reynolds American
Rhodia
Rio Tinto Alcan
Rohm and Hass Company
RWE
Safeway
Sanofi-Aventis
Sara Lee Corporation
Scandinavian Airlines System Denmark-Norway-Sweden
Sempra Energy
Simon Property Group
SOCIETE GENERALE
SOL MELIA, S OCIEDAD ANONIMA
Stagecoach Group Plc
Standard Chartered Bank
Starwood Hotels & Resorts Wordwide
STMicroelectronics N.V.
Stora Enso Oyj
Suez
SUPERVALU Inc.
Svenska Cellulosa Aktiebolaget SCA
Target Corporation
Telecom Italia SPA
Telefonaktiebolaget L M Ericsson
Telenor ASA
Temple-Inland
Textron Financial Corporation
The Allstate Corporation
The Black & Decker Corporation
The Chubb Corporation
The Dow Chemical Company
The Gap Inc.
The Hartford Financial Services Group
The Home Depot
The KROGER Corporation
The Sherwin-Williams Company
The Walt Disney Company
ThyssenKrupp AG
Time Warner
Total SA
Transocean
Tyson Foods
UBS
Unicredit Societa per Azioni
Unilever N.V.
Union Fenosa S.A.
Union Pacific Corporation
United Business Media Plc
UnumProvidentCoporation
UPM-Kymmene Oyj
Valeo
Valero Energy Corporation
Veolia Environnement
Verizon Communications
Vinci
Visteon Corporation
VW
Wal-Mart Stores
Washington Mutal
Weyerhaeuser Company
Whirlpool Corporation
Wyeth
yĞƌŽdžŽƌƉŽƌĂƚŝŽŶ
Country
United States
United States
Spain
United States
France
United States
United States
Germany
United States
France
United States
Nordic
United States
United States
France
Spain
UK
UK
United States
Switzerland
Finland
France
United States
Sweden
United States
Italy
Sweden
Finland
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
Germany
United States
France
United States
United States
Switzerland
Italy
Netherlands
Spain
United States
UK
United States
Finland
France
United States
France
United States
France
United States
Germany
United States
United States
United States
United States
United States
United States
Sector
TMT
Industrials
Energy
Consumer
Industrials
Industrials
Industrials
Energy
Consumer
Consumer
Consumer
Consumer
Energy
Financials
Financials
Consumer
Consumer
Financials
Consumer
TMT
Energy
Energy
Consumer
Consumer
Consumer
TMT
TMT
TMT
Consumer
Financials
Financials
Consumer
Financials
Industrials
Consumer
Financials
Consumer
Consumer
Industrials
TMT
Industrials
TMT
Energy
Energy
Consumer
Financials
Financials
Consumer
Energy
Industrials
TMT
Financials
Industrials
Industrials
Energy
Industrials
TMT
Industrials
Industrials
Industrials
Consumer
Financials
Industrials
Consumer
Consumer
Consumer
Index Group
y
y
iTraxx
y
iTraxx
y
y
iTraxx
y
iTraxx
y
iTraxx
y
y
iTraxx
iTraxx
iTraxx
iTraxx
y
iTraxx
iTraxx
iTraxx
y
iTraxx
y
iTraxx
iTraxx
iTraxx
y
y
y
y
y
y
y
y
y
y
y
y
iTraxx
y
iTraxx
y
y
iTraxx
iTraxx
iTraxx
iTraxx
y
iTraxx
y
iTraxx
iTraxx
y
iTraxx
y
iTraxx
y
iTraxx
y
y
y
y
y
y
XXXVIII
Curriculum Vitae
Curriculum Vitae
Personal Details
Name:
Stefan Morkötter
Date of birth:
5th June 1981
Place of birth:
Hamm, Germany
Education
2007 - 2009
University of St. Gallen
PhD Programme (Dr. oec HSG)
2009
University of Oxford
Visiting PhD Student
2005 - 2007
University of St. Gallen
Master of Arts in Banking and Finance (M.A. HSG)
2006
Tuck School of Business, Dartmouth College
Exchange Term
2004 - 2005
Hogeschool Zeeland
Bachelor of Business Administration (B.B.A.)
2003 - 2005
University of Münster
Studies on Business Administration
2001 - 2005
University of Applied Sciences for Economy and Management
Diplom-Kaufmann (FH)
1991 - 2000
Kardinal-von-Galen Gymnasium
Abitur
Working Experience
since 2007
University of St. Gallen
2008
Shanxi University of Finance and Economics,
Institute of International Studies, China
2007
Deutsche Bank AG, Duesseldorf, Germany
2007
Shanxi University of Finance and Economics,
Institute of International Studies, China
2005
Remondis AG, Moenchengladbach, Germany
2005
Deutsche Bank AG, Duesseldorf, Germany
2004
Deutsche Bank AG, Essen, Germany
2001 - 2003
Deutsche Bank AG, Essen, Germany
2000 - 2001
1. GE/NL Corps, Muenster, Germany