City Research Online

Abstract

This thesis presents three studies in Credit Risk Modelling. The first two studies are exploratory in nature. They investigate whether structural models can be used to price sovereign debt and obtain a creditworthiness variable for countries. In the first study, we test the ability of an extended structural model proposed by Cathead and El-Jahel (2003) to capture the dynamics of the Mexican Brady Par. Using market prices and a Kalman Filter methodology, we estimate the model and obtain the distance-to-default, which is an implicit variable that drives the country’s creditworthiness. The model is slightly superior to one which assumes that the distance-to-default follows a random walk. We find that approximately 80% of the distance-to-default can be explained by just a few economic factors. When this variable is approximated from these factors and substituted back into the models, the Cathead and El-Jahel model still pedorms better than the naive model both in sample and out-of sample, albeit only by a small margin.

The second study extends the above analysis and investigates the dynamics and co movements of the distance-to-default across different countries, specifically: Argentina, Brazil, Mexico and Venezuela. We find that a few country fundamental variables, and external variables, including a variable that measures market sentiment, are able to explain up to approximately 80% of the distance-to-default of each country. Although country specific factors are statistically significant in explaining the distance-to-default, external factors (such as the US stock market index, interest rates and market interdependence across countries) are much more impodant in explaining the dynamics of this variable.

The last study makes a comparison between two Credit Podfolio Risk Models: CreditMetrics versus CreditRisk+. The paper builds on work done by Koyluoglu and Hickman (1998), but we make a significant extension by assessing the impact of migration risk on credit-risk. We make a very careful comparison of Credit-Value-at-Risk for the two models using Monte Carlo techniques on standardised podfolios of bonds. The conclusion is that for regulators, which model is used matters very little. This is because regulators are concerned with extreme values, and loss distributions of both models capture information about defaults at very high confidence levels. However, for internal purposes, where rating migrations matter more than default, CreditMetrics can generate higher estimates of risk.

Publication Type:	Thesis (Doctoral)
Subjects:	F History United States, Canada, Latin America > F1201 Latin America (General) H Social Sciences > HG Finance
Departments:	Bayes Business School > Bayes Business School Doctoral Theses Bayes Business School > Finance Doctoral Theses

Export

Downloads