Opportunistic tactical portfolio management using genetic algorithm and reinforcement learning

Abstract

Market trend reversals are what allow investors to capture profits, but stock trading comes with risks. A good portfolio is therefore one that can diversify risks yet exploit market trend reversals to maximize returns. This paper describes the use of 3 portfolio rebalancing strategies to explore the impact of base rates, commission rate loss and exchange rate loss on a portfolio’s profit level. The three rebalancing strategies are Opportunistic Tactical Buy and Hold Strategy, Genetic Algorithm Rebalancing Strategy (with risk algorithm) and Reinforcement Learning Rebalancing Strategy.

Opportunistic Tactical Buy and Hold Strategy is a rule-based portfolio rebalancing strategy that takes advantage of the relative difference in each index/stock’s country risk to perform rebalancing during trend reversals. It demonstrated the ability to effectively adjust portfolio composition dynamically, as exemplified by experiments performed on both indexes and stocks. It is found that this algorithm performed extremely well in returning a high level of profits, although further tests are needed to verify this result. The effect of base rates is also tested on this strategy, and it is found that the algorithm performed better without base rate. In short, the Opportunistic Tactical Buy and Hold Strategy is able to outperform all underlying buy-and-hold indexes/stocks in each experiment, despite commission rate loss and exchange rate loss. However, a significant amount of profit loss is due to commission rate loss and exchange rate loss. It is found that choosing stocks of highly correlated currency pairs can reduce exchange rate loss, through experiments of portfolios with stocks selected from the telecommunications sector and the healthcare sector.

Genetic Algorithm Rebalancing Strategy (with risk algorithm) is an algorithm that accounts for all three aspects of a market, namely market trends, risks, and returns. It shows promising results, as it is able to adjust the portfolio compositions to maximize returns. It is found that the algorithm performed better with base rates and is able to outperform all underlying buy-and-hold indexes/stocks in each experiment, despite commission rate loss and exchange rate loss.

However, Reinforcement Learning Rebalancing Strategy shows unfavourable results as it was unable to outperform all indexes/stocks. With the introduction of a commission_reward function to account for the significance of commission rate loss, the Enhanced RL rebalancing strategy improved significantly and is able to outperform all underlying buy-and-hold indexes/stocks in terms of total returns, with and without base rate.