Generating data for financial portfolio optimization

This Thesis investigates on the benefits and inconveniences of using data generative models for financial portfolio optimization (FPO). In a first step, we explain the state of the art generative approaches. Then, we focus on a novel approach, until now unemployed to FPO, which is to use generative models that combine time series forecasting models with normalizing flows. We focus on two specific financial strategies, a strategy that maximizes the Sharpe Ratio, and a strategy that minimizes the volatility of a selected portfolio and make comparisons between the multiple approaches and the different models that we choose. Finally we draw conclusions and try to give an unbiased opinion on the results gathered throughout this thesis.