Reinforcement learning has been shown capable of learning optimal strategies from imperfect information environments in order to create robust decision support systems. This paper shows that two automatic feature transformation techniques - Bayesian recurrent neural network (BRNN) for modelling future price trends and Generative Adversarial Networks (GANs) for modelling short-term realistic price variability - are able to improve the performance of reinforcement learning agents in solving portfolio management problem effectively, when measured in terms of increasing profitability and reducing risks.
|Original language||English (US)|
|Number of pages||10|
|State||Published - 2020|
|Event||2020 Spring Simulation Multiconference, SpringSim 2020 - Virtual, Online|
Duration: May 18 2020 → May 21 2020
All Science Journal Classification (ASJC) codes
- Computer Networks and Communications