Dynamic portfolio rebalancing through reinforcement learning

Abstract Portfolio managements in financial markets involve risk management strategies and opportunistic responses to individual trading behaviours. Optimal portfolios constructed aim have a minimal with highest accompanying investment returns, regardless of market conditions. This paper focuses on providing an alternative view maximising portfolio returns using Reinforcement Learning (RL) by considering dynamic risks appropriate conditions through rebalancing. The proposed algorithm is able improve introducing the rebalancing vigorous RL agent. done while accounting for conditions, asset diversifications, global market. Studies been performed this explore four types methods variations fully gradual rebalancing, which combine without use Long Short-Term Memory (LSTM) model predict stock prices adjusting technical indicator centring. Performances evaluated compared three portfolios, including one index assets different levels, two uncorrelated from sectors levels. Observed experiment results, agent LSTM price prediction outperforms other methods, as well these portfolios. improvements are achieved at about 27.9–93.4% over those full model. It has demonstrated ability dynamically adjust compositions according trends, indices assets.