Dynamic conditional correlation models and portfolio risk management

The contribution of this thesis in empirical finance is twofold: On the one hand it delivers a first comprehensive analysis of Dynamic Conditional Correlation (DCC) models, a modern class of risk models. Distinct model specifications, in-sample and out-of-sample forecasts, small up to large sized portfolios of assets and various statistical tests are considered. On the other hand those risk estimates and those of alternative risk models are evaluated in a broad collection of portfolio (risk) management exercises under realistic settings. Alternative portfolio construction techniques/optimizations, various amounts of trading costs and simulated portfolio managers with different skill levels are considered. The key finding is that DCC models do indeed often perform significantly better than other risk models if our objective is to monitor risk. But if we use DCC models in a portfolio optimization context, there exist simple models, like the sample covariance, which perform better indicated by many financial measures. The results of this thesis indicate that active risk control or trading should not be based on DCC models due to much higher portfolio turnover values and consequently much higher trading costs induced by DCC models. However DCC models are recommended if we want to observe or estimate general risk levels in the markets, like standard risk management departments in the practitioner community do as means of reporting. A further important finding is that not only do simple risk models but also simple portfolio constructions often perform at least as well as more complex techniques. Often the most robust portfolios are based on simple techniques, producing best return-risk tradeoffs and best risk-control opportunities. Possible reasons are additional estimation error and higher turnover, hence higher trading cost through portfolio weight instability. But simple constraints which only slightly increase the complexity of the problem, like adding a short selling constraint or an upper bound on tracking error, may indeed also lead to more robust portfolios. In fact, given we choose to consider an optimization-based portfolio construction, it is not recommended to use unconstrained portfolio optimizations. Furthermore the results discourage relying on a single risk approach, as we are not yet able to forecast and control risk to a satisfactory level of accuracy with a single risk concept, particularly not in turbulent market times.