Neural networks are powerful tools for approximating complex functions. In contrast to traditional multivariate distributions, vine copulas provide flexible methods for quantifying the dependence between variables. This thesis investigates the use of feed-forward networks to estimate copula families and their parameters in a time-series context, allowing the dependence to change over time. We assess this approach through extensive simulations and propose a mixture of copulas with a neural gating mechanism, evaluating its performance in a time-indexed setting influenced by seasonal factors. Finally, we apply this method to weather ensemble forecasts.     «

Neural networks are powerful tools for approximating complex functions. In contrast to traditional multivariate distributions, vine copulas provide flexible methods for quantifying the dependence between variables. This thesis investigates the use of feed-forward networks to estimate copula families and their parameters in a time-series context, allowing the dependence to change over time. We assess this approach through extensive simulations and propose a mixture of copulas with a neural gating...     »