DefaultRisk.com the web's biggest credit risk modeling resource. |  |
|
|
| Lévy Subordinator Model: A two parameter model of default dependency by B.S. Balakrishna of the Unaffiliated June 28, 2011 Abstract: The May 2005 crisis and the recent credit crisis have indicated to us that any realistic model of default dependency needs to account for at least two risk factors, firm-specific and catastrophic. Unfortunately, the popular Gaussian copula model has no identifiable support to either of these. In this article, a two parameter model of default dependency based on the Lévy subordinator is presented accounting for these two risk factors. Subordinators are Lévy processes with non-decreasing sample paths. They help ensure that the loss process is non-decreasing leading to a promising class of dynamic models. The simplest subordinator is the Lévy subordinator, a maximally skewed stable process with index of stability 1/2. Interestingly, this simplest subordinator turns out to be the appropriate choice as the basic process in modeling default dependency. Its attractive feature is that it admits a closed form expression for its distribution function. This helps in automatic calibration to individual hazard rate curves and efficient pricing with Fast Fourier Transform techniques. It is structured similar to the one-factor Gaussian copula model and can easily be implemented within the framework of the existing infrastructure. As it turns out, the Gaussian copula model can itself be recast into this framework highlighting its limitations. The model can also be investigated numerically with a Monte Carlo simulation algorithm. It admits a tractable framework of random recovery. It is investigated numerically and the implied base correlations are presented over a wide range of its parameters. The investigation also demonstrates its ability to generate reasonable hedge ratios. Keywords: default risk, correlation smile, CDO, Lévy process, subordinator, semi-analytical, FFT, copula, catastrophe Books Referenced in this paper: (what is this?)
Download paper (667K PDF) 39 pages [ |
