Discussion on A Dynamic Model for the Forward Curve or the CFRS model Authors: Choong Tze Chua, Dean

1
Discussion on A Dynamic Model for the Forward
Curve(or the CFRS model)Authors Choong Tze
Chua, Dean Foster, Krishnaswamy and Robert
Stineby Sandeep Juneja, TIFRWinter Research
Conference Dec 21-22, ISB
2
Paper Summary

• Developed a class of general, arbitrage free,
  forward rate models based on three economically
  sensible dynamics
• Unconditional steady state forward curve
• Maturity specific curve
• Date specific curve
• Developed easy parameterization of the model to
  allow model estimation and computation of bond
  prices and yields
• Empirical results suggest that the model provides
  superior forecast to existing benchmark models

3
Key Strengths

• Based on sound economic rationale
• Superior forecast compared to existing benchmarks
• Strong empirical analysis
• Arbitrage free model has a straightforward
  representation
• Interest rate derivatives are expectations under
  risk neutral measure
• Easy to price them via numerical procedures
• Bond prices and yields can be explicitly computed
  
• Very comprehensive. Addresses all important
  issues
• Well written

4
Comments on the model

• Developed a class of arbitrage free, general
  forward rate models based on three economically
  sensible dynamics
• Unconditional steady state forward curve
• Maturity specific curve
• Date specific curve
• Does the data suggest these dynamics?
• Can observable quantities such as the LIBOR rates
  be modeled in this framework ?

5
Comments on parameterizations

• The models proposed are a specific
  parameterizations of the HJM models.
• Diffusions are carefully selected, then the
  drifts follow from the no-arbitrage condition
• Does there exist a direct methodology for
  applying HJM framework to the forecasting
  problem, to compare with the proposed method?

6
Comments on Experiments

• Out-of-sample forecasts superior to those of
  available benchmark models
• Can contributions to accuracy by each of the
  three model components be segregated?
• Modifications to improve short time horizon
  predictions
• Model is selected as the best amongst proposed
  seven. Why these seven? Are these best in any
  sense?

7
Further Comments

• Once the unconditional curve is estimated, only
  two parameters are estimated from data.
• Any comments on this on the assumption of
  homoscedasticity in innovations.
• Distribution of forward rates and spot rates in
  CFRS model?
• Useful in pricing interest rate derivatives using
  Monte-Carlo

8
Last Point

• Interesting Comparison
• Same data
• Different models
• Compare price of interest rate derivatives