Scholarlink

<p>Predicting market movements and developing successful trading techniques is difficult due to the nonlinear dynamics and extreme volatility of financial markets. Conventional machine learning (ML) and deep learning (DL) methods, including LSTM (Long Short-Term Memory) and CNN (Convolutional Neural Networks), struggle with real-time decision-making under uncertainty. Traditional supervised forecasting approaches provide price direction signals but lack adaptability in dynamic markets. The objective is to improve the efficiency and precision of predictions and trades. The objective is to improve the efficiency and precision of predictions and trades by integrating deep reinforcement learning (DRL) with stable multi-agent decision optimization. A framework named Multi-Agent Proximal Policy Optimization with Stacked LSTM (MAPPO-SLSTM) is proposed, combining temporal sequence modeling with cooperative reinforcement learning to capture cross-asset dependencies. MAPPO-SLSTM employs a stacked LSTM architecture to encode sequential price signals and a multi-agent PPO scheme with a centralized critic to optimize portfolio-level actions in volatile environments. Historical OHLCV (Open, High, Low, Close, Volume) cryptocurrency data through minute-level granularity is utilized, ensuring sufficient interaction samples for reinforcement learning and convergence. Missing value imputation and Min–Max scaling are applied to normalize market data. Technical indicators, log returns, volatility measures, and dimensionality reduction through Principal Component Analysis (PCA) are incorporated to enrich market states. LSTM layers extract temporal features, the centralized critic evaluates joint policies, and decentralized actors execute optimized buy–hold–sell strategies across multiple assets. The results are evaluated using Python 3.10, with performance assessed through an accuracy of 0.946, and other metrics. The integration of SLSTM with multi-agent PPO establishes a robust framework for adaptive financial trading in highly volatile markets.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Financial asset price prediction and quantitative trading strategy optimization based on deep reinforcement learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月29日

查看全文
  • Shuxin Li

摘要

Predicting market movements and developing successful trading techniques is difficult due to the nonlinear dynamics and extreme volatility of financial markets. Conventional machine learning (ML) and deep learning (DL) methods, including LSTM (Long Short-Term Memory) and CNN (Convolutional Neural Networks), struggle with real-time decision-making under uncertainty. Traditional supervised forecasting approaches provide price direction signals but lack adaptability in dynamic markets. The objective is to improve the efficiency and precision of predictions and trades. The objective is to improve the efficiency and precision of predictions and trades by integrating deep reinforcement learning (DRL) with stable multi-agent decision optimization. A framework named Multi-Agent Proximal Policy Optimization with Stacked LSTM (MAPPO-SLSTM) is proposed, combining temporal sequence modeling with cooperative reinforcement learning to capture cross-asset dependencies. MAPPO-SLSTM employs a stacked LSTM architecture to encode sequential price signals and a multi-agent PPO scheme with a centralized critic to optimize portfolio-level actions in volatile environments. Historical OHLCV (Open, High, Low, Close, Volume) cryptocurrency data through minute-level granularity is utilized, ensuring sufficient interaction samples for reinforcement learning and convergence. Missing value imputation and Min–Max scaling are applied to normalize market data. Technical indicators, log returns, volatility measures, and dimensionality reduction through Principal Component Analysis (PCA) are incorporated to enrich market states. LSTM layers extract temporal features, the centralized critic evaluates joint policies, and decentralized actors execute optimized buy–hold–sell strategies across multiple assets. The results are evaluated using Python 3.10, with performance assessed through an accuracy of 0.946, and other metrics. The integration of SLSTM with multi-agent PPO establishes a robust framework for adaptive financial trading in highly volatile markets.