Reinforced learning for portfolio management

Abstract

The use of reinforcement learning in managing portfolios is a current area of focus in the financial technology field. This research aims to find the best way to redistribute a fund among different financial assets over an extended period, through trial and error. Current methods have limitations, as they typically assume that each redistribution can be completed immediately, ignoring the impact of price changes as a cost of trading. To address these issues, a proposed solution is a hierarchical system for managing portfolios using reinforcement learning (HRPM). Main contribution from the author is building a full-scale front-end website for the organisation, TradeMaster. Another contribution is assisting in testing of the backend algorithms. This report will discuss about factors that is fundamental to a good working frontend website and the fundamentals of reinforced learning in stocking trading. It will also show the implementation of the website and the results of the algorithm testing.