Understanding joint distributions of asset returns is a key challenge for financial risk management. In this master thesis we conceptualise, implement and study a novel copula-based approach to conditional multivariate risk modelling, as introduced in the preliminary paper by (1). Contrary to classical copula estimation methodologies, which separately estimate the marginal distribution and dependence structure in two steps, our approach admits the joint learning of model parameters from the conditional features via a neural network specification. To accomplish this in a scalable manner, we employ a convenient factorisation of the correlation matrix, which informs the copula specification. The main goals of this master project entail the introduction of the modelling framework and to develop a first operational implementation, capable of estimating the model parameters, with the aim of testing its performance on data and discussing potential applications. We show that the model can be fitted to data adequately, however its generalisation to out-of-sample observations is limited as of yet. In particular the estimation of conditional mean returns proves to be difficult. On the other hand, we show the viability of the model for use in portfolio optimisation (mean-variance investment) and value-at-risk calculations. (1) Damir Filipović and Puneet Pasricha. Copula process models for financial risk management (preliminary version). 2022.      «

Understanding joint distributions of asset returns is a key challenge for financial risk management. In this master thesis we conceptualise, implement and study a novel copula-based approach to conditional multivariate risk modelling, as introduced in the preliminary paper by (1). Contrary to classical copula estimation methodologies, which separately estimate the marginal distribution and dependence structure in two steps, our approach admits the joint learning of model parameters from the cond...     »