Evaluating Credit Default Swap Spreads: A Comparative Study of Traditional Methods and an Investigation of their Assumptions

Sammanfattning

Due to the global financial crisis 2008, the requirements for calculating and valuing default risk in financial derivatives have tightened and are now regulatory-guided worldwide. Measurements of default risk in financial derivatives are performed through Counterparty Credit Risks. Credit Value Adjustments (CVA) are used to price financial derivatives by incorporating the default risk. In order to calculate CVA, Credit Default Swaps (CDS) spreads play a key role in evaluating the risk-neutral pricing due to counterparty risk. Traditionally, there are two standard methods. The most straightforward approach is to use the intersection method. This method averages liquid CDS quotes of parties within the same industry sector and geographical region with the same credit rating. This is done assuming these parties are expected to behave similarly regarding default risk. The NOMURA model, or cross-section method, bases its approximations on the same assumptions but instead uses multi-dimensional linear regression to generate the proxy models of the CDS spreads. These assumptions are rooted in credit risk modeling theory and market practice. In other words, two parties within the same subcategory are expected to have similar credit profiles. In practice, basing the approximations on these assumptions greatly affects the outcome. Therefore, this thesis not only compares the two methods separately but also investigates the assumptions and whether the behavior of the liquid CDS can be traced back to the expected behavior quoted historically. This thesis evaluation of the assumptions is performed by hierarchical clustering of the obligors with liquid CDS spreads based on their CDS quotes historically. The distribution of obligors over the clusters indicates the reasonability of the assumptions. This project showed little difference between the two methods' proxies generated using this data set. Moreover, the cluster analysis struggled with meaningfully clustering the data. There are no clear distinctions between the subcategories, which is assumed. Similar credit ratings tend to behave similarly. Generally, the conclusions that can be drawn from the comparison and the cluster analysis are rather vague. However, conducting a cluster analysis on a larger data set would be interesting to examine the assumptions in practice further.