Novel Deep Learning Approaches to Portfolio Allocation and Risk Management

The first part of my research focuses on the review of heuristic portfolio optimisation techniques by carrying out a number of empirical out of sample studies on several cross-asset universes. The results of this performance evaluation are then used towards the main research goal, which is the application of machine learning techniques to identify market regimes where a particular asset allocation strateguy is closest to the mean-variance optimum, as described by Harry Markowitz. Latest research in empirical finance, information theory and biodiversity accumulated a list of mathematical tools that can now be used to accurately describe leading characteristics of an asset universe at each point in time. I am particularly interested in studyingthe effectiveness of deep learning in detecting and exploiting non-linear interactions among these metrics and whether they possess enough forecasting power to accurately predict relative performance of each portfolio allocation strategy. Finally, my research proposes a novel portfolio allocation procedure, which dynamically selects a strategy with the best risk-adjusted performance in the future investment period, taking into account transaction costs.