Stock Portfolio Management by Using Fuzzy Ensemble Deep Reinforcement Learning Algorithm

The research objective of this article is to train a computer (agent) with market information data so it can learn trading strategies and beat the index in stock without having make any prediction on moves. approach assumes no knowledge, agent will only from conducting historical data. In work, we address task by considering Reinforcement Learning (RL) algorithms for portfolio management. We first generate three-dimension fuzzy vector describe current trend each stock. Then terms, along other features, such as prices, volumes, technical indicators, were used input five algorithms, including Advantage Actor-Critic, Trust Region Policy Optimization, Proximal Actor-Critic Using Kronecker Factored Region, Deep Deterministic Gradient. An average ensemble method was applied obtain actions. set SP100 component stocks pool 11 years daily model simulate trading. Our demonstrated better performance than two benchmark methods individual algorithm extension. practice, real traders could use trained inferences conduct trading, then retrain once while since training models time0consuming but making nearly simultaneous.