ISLAMIC STOCK PORTFOLIO OPTIMIZATION USING DEEP REINFORCEMENT LEARNING

Abstract

The Islamic principles in identifying stocks as Shari’ah principles have inevitability restrict the number of stocks that Muslims can invest in and consequently may affect the return from investment. In this paper, we examine the potential of Deep Reinforcement Learning in optimizing the portfolio returns of Islamic stocks. We model stock trading as a Markov Decision Process problem because of its stochastic and interactive nature. Then, we define the trading objective as a problem of maximization, while the DRL agents used are actor-critic algorithms. The selected portfolio consists of 30 most liquid Islamic stocks in Indonesia that constitute JII index and compare with that of the benchmark portfolio, namely the 45 most liquid conventional stocks or LQ45. The performance is compared using several algorithms. The result show that trading on Islamic stocks from January 2019 to December 2020 using the DRL agents could outperform the benchmark index of conventional stocks. Using DRL agents, fund managers would be able to optimize the portfolio on daily basis, minimize risk during crisis or turbulence, and outperform the conventional stocks.

Keywords

• deep reinforcement learning, actor-critic framework, islamic stock.