Deep graph convolutional reinforcement learning for financial portfolio management – DeepPocket

Portfolio management aims at maximizing the return on investment while minimizing risk by continuously reallocating assets forming portfolio. These are not independent but correlated during a short time period. A graph convolutional reinforcement learning framework called DeepPocket is proposed whose objective to exploit time-varying interrelations between financial instruments. represented nodes correspond instruments edges pair-wise correlation function in assets. consists of restricted, stacked autoencoder for feature extraction, network collect underlying local information shared among instruments, and an actor-critic agent. The structure contains two networks which actor learns enforces policy is, turn, evaluated critic order determine best course action constantly various portfolio optimize expected investment. agent initially trained offline with online stochastic batching historical data. As new data become available, it passive concept drift approach handle unexpected changes their distributions. against five real-life datasets over three distinct periods, including Covid-19 crisis, clearly outperformed market indexes.