Die Kapitalstruktur europäischer Emittenten von bedingten Pflichtwandelanleihen im Bankensektor: eine robuste Korrelationsund Portfoliooptimierungsanalyse mit besonderer Berücksichtigung der Wavelet-Analyse und von Bayesschen Netzen

This thesis investigates alternative estimation procedures of portfolio parameters to alleviate the effects of structural breaks, whereby the portfolio comprises capital structure instruments of European banks that have contingent convertible bonds outstanding. The robust estimation procedures focus on the correlation of returns and are based on the Wavelet Analysis and on Bayesian Nets. The estimation results will feature rolling and extending sample windows and be implemented in a standard mean-variance optimisation algorithm to construct optimal portfolios. The performance of these optimal portfolios will then in turn be further analysed and compared to the performance of reference portfolios with traditionally estimated parameters. The performance analysis will be carried out with asymmetric and higher-order measures to get a more authentic risk notion and to deal with non-normality of the return distribution.

The estimation procedures based on the Wavelet Analysis will conduct a Multi Resolution Analysis through the Maximum Overlap Discrete Wavelet Transformation. This will enable the calculation of different correlation measures on varying time scales. The traditional Pearson Correlation and the Dynamic Conditional Correlation are used for the correlation calculations of the transformed data. The reference portfolios will only use original data. The comparing study between the performance of the reference portfolios and the portfolios estimated with the transformed data show that parameters estimated at different time scales can have a positive impact on risk-adjusted portfolios returns. Especially, lower-frequency correlations unveil favourable diversification effects which were left undetected by traditional methods.

The estimation procedure based on Bayesian Nets models the causal relationships, which trigger a structural change of the return distribution of capital instruments associated with a stressed market state. On the foundation of the constructed Bayesian Net and the historical distribution in normal times, a joint probability distribution will be calculated. This joint probability distribution will be used to estimate the required portfolio parameters. The results are compared to traditional methods, whereby they show that risk-adjusted returns are improved and the alternative approach is coping better with a stressed market environment.