DefaultRisk.com the web's biggest credit risk modeling resource. | 
|
|
|
| Explaining Base Correlation Skew Using NG (Normal-Gamma) Process by Siddharth Hooda of Nomura International, Plc June 6, 2006 Abstract: In order to explain the base correlation skew effect, practitioners have been trying non-Gaussian based one factor copula model. This has given rise to a new series of model based on Levy process (Joshi-Stacey) [6], Variance Gamma process (Moosbrucker)[7], B-VG process (Baxter)[2], NIG (Guegan and Houdain) [4] and One Factor Levy Processes (Schoutens) [1]. All these models, introduce longer and fatter tails for global and idiosyncratic factors by using Levy processes. Similar on line with these non-Gaussian based copula models, we have introduced a copula based on a new process called Normal-Gamma (NG). NG differs from the other models in terms of source of jump information. NG model assumes that there is only one universal source of jump process as compared to correlated jump processes. Both global as well as idiosyncratic factors react to the universal jump process at the same time but may react in different direction as well as with a different magnitude. We employ Hull-White [5] recursive algorithm for building portfolio loss distribution and consequently evaluating Tranches Par Spreads. NG copula explains the Base Correlation Skew better than the Gaussian Factor copula model. NG copula inferred Base Correlation curve is less skewed than the Gaussian copula implied Base Correlation curve and hence much easier to interpolate. Calibration results with Average Absolute Error (AAE) and Average Relative Absolute Error (ARAE) has been illustrated in paper. Keywords: CDO Pricing, Gamma Process, Base Correlation Skew, Loss Distribution, Copula, Tranchelets and Calibration. Books Referenced in this Paper: (what is this?)
Download paper (939K PDF) 22 pages
[Home] [Credit Correlation Papers]  |   
|
Please contact me with problems or suggestions. |
