Reinforcement Learning-Driven Portfolio Management: Adaptive Strategies for Risk and Returns Using LSTM and DQN

The goal of portfolio optimization in investment management deals with combining return objectives with risk control under conditionally changing and volatile market environments. Standard optimization procedures fail to capture market dynamics alongside financial market non-stationarity patterns since they result in inferior decisions. Building a system of continuous learning models that makes data-driven decisions in real time represents the main challenge in this task. The proposed solution meets these challenges through an RL algorithm integration with LSTM networks for stock price forecasting accuracy and Deep Q-Networks (DQN) implemented for instant Buy, Hold or Sell decisions. The model achieves portfolio return maximization by minimizing risk through adaptive performance that tracks market changes and improves its decisions continuously. The hybrid model functions better than traditional systems in handling market fluctuations because it simultaneously advances portfolio returns while managing real-time risk proportions effectively. Based on its R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value of 0.78 the LSTM model demonstrates adequate performance in detecting stock price connections but the DQN model relies on cumulative reward to optimize trading decisions.