Scholarlink

<p>In recent decades, Artificial Intelligence (AI) has been developed to enhance performance in various sectors. In the field of portfolio optimization, it is crucial to balance investment strategies with the aim of high returns and low risk. Algorithmic trading has improved enormously with the help of Deep Reinforcement Learning (DRL), which is more reliable for portfolio optimization with a dynamic number of assets. This research endeavours to provide a broad view of DRL in trade, improving trading choices. Five kinds of DRL agents are proposed to optimize portfolios: stock selection agent, trend prediction agent, risk management agent, trade execution agent and portfolio rebalancing agent. In this portfolio optimization, the stock selection process is performed using Binary Genetic Siberian Tiger Optimization and decision trees (BGSTO-DC) to select an optimal subset of stocks from the ALGO TRADING NIFTY-500 dataset. After stock selection, the Agentic AI model, namely the dilated LSTM-based Transformer model with a progressive attention mechanism, is employed to predict the trends. After analysing risks, Deep Q-Network (DQN) based RL is introduced as a trade execution agent. Based on the actions attained from DQN, the trading takes place while optimizing transaction costs. The Markowitz Modern Portfolio Theory (MPT) model is used to rebalance the portfolio to maximize returns while managing risk dynamically. The outcomes of this research indicate that the presented model outperforms all other existing frameworks by generating a high annual return and cumulative return and Sharpe ratio. The results highlight its appropriateness with enhanced trading strategies.</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

Agentic AI-driven portfolio optimization: a hybrid approach for optimized stock selection and deep learning in algorithmic trading

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Sudheer Singamsetty,
  • Satyanarayana Sanakkayala

摘要

In recent decades, Artificial Intelligence (AI) has been developed to enhance performance in various sectors. In the field of portfolio optimization, it is crucial to balance investment strategies with the aim of high returns and low risk. Algorithmic trading has improved enormously with the help of Deep Reinforcement Learning (DRL), which is more reliable for portfolio optimization with a dynamic number of assets. This research endeavours to provide a broad view of DRL in trade, improving trading choices. Five kinds of DRL agents are proposed to optimize portfolios: stock selection agent, trend prediction agent, risk management agent, trade execution agent and portfolio rebalancing agent. In this portfolio optimization, the stock selection process is performed using Binary Genetic Siberian Tiger Optimization and decision trees (BGSTO-DC) to select an optimal subset of stocks from the ALGO TRADING NIFTY-500 dataset. After stock selection, the Agentic AI model, namely the dilated LSTM-based Transformer model with a progressive attention mechanism, is employed to predict the trends. After analysing risks, Deep Q-Network (DQN) based RL is introduced as a trade execution agent. Based on the actions attained from DQN, the trading takes place while optimizing transaction costs. The Markowitz Modern Portfolio Theory (MPT) model is used to rebalance the portfolio to maximize returns while managing risk dynamically. The outcomes of this research indicate that the presented model outperforms all other existing frameworks by generating a high annual return and cumulative return and Sharpe ratio. The results highlight its appropriateness with enhanced trading strategies.