Accurate estimation of risk measures for financial portfolios is of great importance equally for financial institutions as well as regulators. Many existing methods lack the ability to adequately incorporate the high dimensional dependence structure of the financial portfolio. In this research we capture the cross dependence of the assets using the flexible class of R-vine copulas and their trend and volatility univariately with ARMA-GARCH models. Given these two components we simulate portfolio level forecasts and follow a Monte Carlo approach to estimate various risk measures on the portfolio level. All of this is performed in a rolling window fashion. This approach extends the work of Maarouf (2021) as not only the Value at Risk (VaR) but also the expected shortfall (ES), which is the successor of the VaR as the primary market risk measure as of the Basel III accords, is considered. A detailed discussion of applicable backtesting strategies for the VaR and ES estimates is also provided. Moreover, in this thesis we introduce a conditional estimation approach. We specify one or two additional market indices or other main market players that are also univariately modelled via ARMA-GARCH models and then incorporated in the D-vine copula which is used in the conditional setting for cross dependence modelling. Then we simulate the portfolio level forecasts conditionally on the market index or indices which leads to conditional portfolio level risk estimates. We then introduce a quantile-based approach to observe the behavior of the risk measures given a certain state of the conditioning asset or assets. In particular, this can generate valuable insights in stress testing situations. An important part of the conditional approach is the conditional sampling from a D-vine copula. Thus, in order to facilitate this approach this thesis also introduces algorithms to sample from the rightmost or the two rightmost leafs of a D-vine copula. The last part of the thesis covers multiple case studies on a Spanish stock portfolio where we successfully apply all the presented methods. Additionally, the algorithms were optimized for computational efficiency. The complete code is publicly available and the developed R package portvine provides efficient implementations for all the risk measure estimation approaches proposed in this thesis.      «

Accurate estimation of risk measures for financial portfolios is of great importance equally for financial institutions as well as regulators. Many existing methods lack the ability to adequately incorporate the high dimensional dependence structure of the financial portfolio. In this research we capture the cross dependence of the assets using the flexible class of R-vine copulas and their trend and volatility univariately with ARMA-GARCH models. Given these two components we simulate portfoli...     »