Henrik Brunlid och Oskar Arnoldsson, presenterar sitt examensarbete


CDO Pricing Using the Normal Inverse Gaussian Copula
and Fast Fourier Transforms  
  	
Abstract

The use of the Gaussian copula has up to this point been the
standard industry model to price synthetic CDO tranches. However, as
have been widely documented in the literature, this model produces a
distinct default correlation smile when calibrated to market CDO
data. This fact contradicts the fundamental model assumption of a flat
default correlation and suggests that the Gaussian copula does not
adequately account for the dependence between extreme credit
events. To counter this problem we suggest the use of a normal inverse
Gaussian (NIG) factor copula. This approach increases the complexity
of the model but it gets rid of the reported correlation smile. In
particular, our results strongly indicate that both skewness and
kurtosis are important characteristics of our factor copula model. In
addition, a complete framework for calculating portfolio loss
distributions using both fast Fourier transform (FFT ) and large
homogenous portfolio (LHP) approximations is presented. We show that
the FFT method yields more stable results than the corresponding LHP
approximation.