Fundamentals-Based versus Market-Based Cross-Sectional Models of CDS Spreads

1
Fundamentals-Based versus Market-Based
Cross-Sectional Models of CDS Spreads

• by
• S. Das, P. Hanouna and A. Sarin
• Discussed by
• J. Helwege
• FDIC September 2006

2
Summary

• Uses a panel dataset of CDS spreads to evaluate
  which factors determine the magnitude of credit
  spreads
• Cross-section but also time-series
• Runs a horse race between accounting data
  determinants of spreads and market-based
  variables.
• Important to evaluate usefulness of accounting if
  private firms
• Structural models tend to ignore accounting data,
  which may be a mistake if it has a lot of
  explanatory power

3
Motivation

• Does this help answer the question of what
  determines yield spreads?
• Is this a better alternative to studies such as
  Collin-Dufresne, Goldstein and Martin or Elton,
  Gruber, Aggarwal, and Mann?
• Put more emphasis on liquidity?
• Longstaff vs. Sundaresan on liquidity issues in
  CDS
• No comparison to old papers that ask what
  determines yield spreads on corporate bonds
  (e.g., Fisher 1959)
• Argument that ranking in the cross section is all
  one needs for convergence trades
• Can we do more with this?

4
Accounting vs market

• Which distinction is more relevant?
• market data vs. book data
• Structural model vs Altman type prediction of
  probability of default
• If former, volatility of operating earnings for
  the industry is a good replacement variable for
  equity volatility (see Helwege and Liang, JFE)
• If latter, want option inputs, esp. vol, separate
  from accounting vars

5
Estimation issues

• In any unbalanced panel, have to ask whether the
  sample is a random, representative sample
• Can get long time series on some firms and not on
  others
• Is existence in the dataset random?
• Use fixed effects
• Are multiple obs in the dataset giving a fair
  sense of weight of a firm?
• With corp fin get about 20 obs for each firm, so
  weights are fairly even
• With stocks, might even toss out of dataset if
  not at least 60 obs per firm
• If more obs in time series, is it liquidity? If
  so, create a variable for the number of times it
  shows up?

6
Estimation issues

• With bonds, there are multiple obs on the same
  firm at a give point in time, depending on the
  firms capital structure
• Need to weight data
• With CDS, can also get variation by maturity of
  contract with same underlying collateral?
• In corp bond lit, use three schemes (see Warga
  and Welch 1993)
• Use all the data
• One bond per firm, preferably a representative
  one
• Average features of bonds for a firm, put in one
  ob
• Table 10 helps by using only 5 year contracts

7
Rankings

• To determine whether accounting or market
  variables better explain the ranking of CDS
  spreads, the authors use CAP curves.
• What is disadvantage of Wilcoxon rank sum tests?

8
Exposition

• When can we get rid of these kinds of sentences?
• The growth of the credit derivatives market
  since the turn of the century has been
  astounding. The OCC reported credit derivative
  volumes of 287 billion at the end of 1999.
  Various estimates now put this volume at over 15
  trillion.
• Credit Default Swaps are contingent claims with
  payoffs that are linked to the credit risk of a
  given entity.
• A CDS is a default insurance contract

9
What I especially like

• Gets back to credit risk instead of trying to say
  that yield spreads are all about liquidity
• Finds an important role for accounting data
• Further proof that the Merton models probability
  of default and KMVs EDF are not sufficient
  statistics
• Gives us confidence that we can do something with
  private firms credit risk
• Puts a large weight on the ability to rank a
  group of credit risky products maybe the best
  way to approach the analysis if we cannot get a
  handle on liquidity premia