Dynamic Portfolio Optimization with Deep Reinforcement Learning: Evidence from Borsa Istanbul

In this study, portfolio optimization has been conducted using the reinforcement learning approach, one of the artificial intelligence algorithms. The data is considered for constituents of the BIST30 index, which is the blue-chip index of Borsa Istanbul. The performance of Deep Deterministic Policy Gradient (DDPG), a deep learning algorithm of reinforcement learning, has been tested against the Markowitz mean-variance and equal-weighted portfolios as benchmark models; the BIST30 index itself has also been taken as a benchmark portfolio. This study contributes to the relevant literature in terms of Türkiye as an example of a developing country and the method employed. The study demonstrates the potential of RL approaches that are becoming widespread for portfolio optimization. The obtained results reveal that the portfolio formed with the DDPG approach shows a superior Sharpe ratio portfolio over portfolios obtained with other classical approaches. These findings, while highlighting the potential of RL approaches in practice, emerge as an alternative option for fund managers, especially in a volatile market environment.