Abstract

This work focuses on the modelling of two specific risk measures, namely the term structure of government bonds and firm credit ratings. We discuss estimation and forecasting issues relating to these risk measures and their relationship to macroeconomic trends. Risk is an important component in any financial decision. Interest rates and credit ratings are two economic variables that reflect financial risk in different ways and are both interrelated. The term structure of interest rates — different rates for different maturities— represent a market perspective while credit ratings represent the view of a credit rating agency on the default probability of an economic entity. Government bond interest rates determine the ability of a country to finance itself and are important factors in the fixture of other interest rates. Modelling the term structure helps in understanding this economic base variable and determine financial risk. We choose a data driven approach in which unseen dynamic factors of the term structure are estimated via principal components analysis in rolling time windows to produce yield curve forecasts. A statistical and economic evaluation is provided for different sets of predictors, estimation methods, and forecasting methods. The data consists of daily observations of government bond interest rates for Germany, Switzerland, the UK, and the US for the time period from 2000 to 2016. Implicitly this approach tests the basic assumptions of Nelson-Siegel type economic factor models of the term structure. Term structure forecasts are evaluated in terms of three complementary criteria or loss functions, namely the statistical mean squared forecast error criterion, and the two more economic criteria of directional accuracy and big hit ability. Factor analysis supports the idea that a level, slope, and curvature factor underly the yield curve. In a data set with all term structures we find evidence of a global level factor. A comparison to simple forecasts such as random walk and autoregressive forecasts shows that dynamic factor models can, in rare instances, improve on random walk forecasts and consistently outperform auto-regressive forecasts under both statistical and economic evaluation criteria. Statistical and economic criteria suggest that more than one factor should be employed for forecasting. Using additional term structures for factor extraction can improve forecasts for some countries depending on the forecast horizon. With regard to estimation methods the standard principal components method using ordinary least squares outperforms the alternative method using generalised least squares. The forecasting method employing autoregressive factors outperforms the method exploiting the lagged correlation of factors and interest rates. These results support the concepts of of the Nelson-Siegel Model. Corporate credit ratings are the traditional business of credit rating agencies. Credit ratings have important effects on financial markets and are a part of financial market regulation. Credit ratings agencies also provide assessments of the credit quality of other entities like countries and structured finance products. The financing of ratings, the competition on rating markets, and the power of debt issuers play an important role in the quality of credit ratings. There are numerous statistical methods to estimate corporate credit ratings. Here, we employ the ordered probit and an unordered logit approch, as well as an OLS approach developed here, that replaces ratings with their respective default rates. Methodologically this approach provides a way to integrate an estimate of the assumed continuous variable that underlies the probit and logit methods, which is unobservable in these models. Furthermore, this approach underscores the connection between credit ratings and default probability. This thesis uses ratings of selected US, UK, German, French, Japanese, Canadian, and Australian firms from 1990 - 2009 and their respective accounting data for corporate credit rating estimation. Here, previous findings are confirmed that show that credit rating agency standards have become more stringent over time given the same accounting data. Furthermore, the results shown here suggests that market pressure outside the US rating market can influence credit ratings agencies judgement.