Pretorius, Ruan2023-05-082023-05-082022https://hdl.handle.net/10539/35474A research report submitted in partial fulfilment of the requirements for the degree of Master of Science to the Faculty of Science, School of Computer Science and Applied Mathematics University of the Witwatersrand, Johannesburg, 2022Traditional portfolio management methods can incorporate specific in- vestor preferences but rely on accurate forecasts of asset returns and co- variances. Reinforcement learning (RL) methods do not rely on these ex- plicit forecasts and are better suited for multi-stage decision processes. To address limitations of the evaluated research, experiments were con- ducted on three markets in different economies with different overall trends. By incorporating specific investor preferences into the proposed RL models’ reward functions, a more comprehensive comparison could be made to traditional methods in risk-return space. Transaction costs were also modelled more realistically by including non-linear changes introduced by market volatility and trading volume. The results of this study suggest that there can be an advantage to using RL methods com- pared to traditional convex mean-variance optimisation methods under certain market conditions. The proposed RL models could significantly outperform traditional single-period optimisation (SPO) and multi-period optimisation (MPO) models in upward trending markets, but only up to specific risk limits. In sideways trending markets, the performance of SPO and MPO models could be closely matched by the proposed RL models for the majority of the excess risk range tested. The specific mar- ket conditions under which these models could outperform each other highlight the importance of a more comprehensive comparison of Pareto optimal frontiers in risk-return space. These frontiers give investors a more granular view of which models might provide better performance for their specific risk tolerance or return targets.enThesis