FIRN Doctoral Tutorial

1
Testing Consumption-based Asset Pricing Models in
Australia
Bin Li, University of Queensland

• FIRN Doctoral Tutorial
• Sydney, December 11, 2007
• Discussant Petko S. Kalev

2
Summary of the Paper

• The study compares four different specifications
  of the CCAPM
• Classic CCAPM by Hansen and Singleton (1982),
  (HS)
• Recursive Utility Model by Epstein and Zin
  (1991), (EZ)
• Habit Formation Model 1 by Abel (1990), (AL)
• Habit Formation Model 2 by Campbell and Cochrane
  (1999), (CC)

3
Contributions

• Adds to the scarce empirical literature of
  estimating/testing CCAPM with Australian data
• Better data utilized (aggregate consumption data
  in Faff, 1998 and Faff Oliver, 1998)
• Longer sample period (19742 - 19924 in both
  Faff, 1998 and Faff Oliver, 1998)
• The paper is logical, methodologically sound,
  also well-structured and well-written
• In summary, this is a publishable work

4
Main Findings

• There is some support for the traditional CCAPM
  (HS, 1982) using AU data
• the equity premium puzzle can be explained
  with it
• Habit models provide more plausible estimates
  of Risk Aversion coefficients
• Habit Formation Model 1 (AL, 1990) improves the
  performance of the CCAPM at both industry and
  economy levels
• Habit Formation Model 2 (CC, 1999) results in
  the lowest coefficient of relative risk aversion
• EZ recursive utility model is the worst
  performing one with AU data

5
A Step Back to the Theory 1

• Price E discount factor x future payoff
• The discount factor (marginal rate of
  substitution) depends on the marginal utility of
  the investor
• Traditional CCAPM assumes a power utility
  function
• This approach leads to three puzzles
• (i) Equity Premium Puzzle (Mehra and Prescott,
  1985)
• (ii) Stock Market Volatility Puzzle (Campbell,
  1999)
• (iii) Risk Free Rate Puzzle (Weil, 1989)

6
A Step Back to the Theory 2

• EZs (1991) utility function can be useful in
  resolving the Risk Free Rate Puzzle but not the
  Equity Premium Puzzle (gamma not high enough)
• In external habit models like AL (1990) and CC
  (1999), habit depends on aggregate consumption
  that is unaffected by agents decisions
• CC (1999) does resolve the Stock Market
  Volatility Puzzle but cannot resolve the Equity
  Premium Puzzle

7
Suggestions/Comments (1)

• Why this study is an important empirical study?
• Can we further motivate the paper?

8
Suggestions/Comments (2)

• Any a-priory expectations?
• What does the THEORY predict?
• Which particular CCAPM empirical model is
  expected to explain the documented Puzzles (one
  or more than one Puzzle at the time?)

9
Suggestions/Comments (3)

• How did you select the four specifications of the
  CCAPM?
• Why not include more (less) specifications of the
  CCAPM for both testing and model comparison?
• How about comparing the traditional CAPM against
  the CCAPM? If not Why not?

10
Suggestions/Comments (4)

• Econometrics
• Can you justify your usage of the
  continuously-updating GMM approach as developed
  by Hansen, Heaton and Yaron (1996) (HHY)?
• Hint Both the two-stage and the iterative GMM
  estimators exhibit poor small sample properties
  (HHY, 1996)
• Warning HHY (1996) caution that their GMM
  estimator may exhibit poor large sample problem.
  The continuously updating estimator is often
  approximately median unbiased but sometimes
  exhibits extreme outlier behaviour
• Remedy see the next slide

11
Suggestions/Comments (5)

• Econometrics 2
• Can you perform some tests for structural
  stability?
• Hint 1 If we have a known breaking point then it
  is only desired to test for instability at this
  point alone (Ghysels and Hall, 1990, JE, Vol. 45
  pp. 121-139)
• Hint 2 If the breaking point is unknown, then
  the null is the broader hypothesis that there is
  no instability at any point in the sample
• A general framework for testing these (new
  optimal tests) is provided by Andrews and
  Ploberger (Econ. 1994, Vol. 62, pp. 1383-1414)
  within the ML estimators
• Sowell (Econ. 1996, Vol 64, pp 1085-1108)
  generalizes these to the GMM estimators

12
Suggestions/Comments (6)

• Minor comments
• On page 15 while reporting the results from Table
  2 The model seems to be rejected only for bond
  returns.
• What about the TBILL?
• On page 17 In Campbell and Cochrane (1999)
  model, gamma is not the risk aversion coefficient
• Local coeff. of risk aversion is gamma/s(t),
  where s(t) is the surplus consumption ratio
• S(t) (c(t)-X(t))/c(t)

13
END