Scholarlink

<p>In this paper, a hybrid method of equity market analysis with deep learning for volatility forecasting and reinforcement learning for portfolio optimization is presented. Volatility forecasting has been performed using Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models on past S&amp;P 500 and VIX index data (2015–2024). GRU performs better than LSTM in terms of the forecast with an R<sup>2</sup> of 0.91. These predictions are aggregated in a trading platform by a Proximal Policy Optimization (PPO) agent that learns to adjust trading strategies from dynamic market indicators like realized volatility, RSI, and MACD. The PPO agent performs outstandingly in portfolio appreciation, increasing capital from $10,000 to over $22,000 with a modest 28.9% winning rate. There is a huge risk-reward ratio and solid decision-making due to it. The research is adequately supported by the appropriateness of sequence models and reinforcement learning towards designing adaptive, intelligent trading algorithms that are highly generalizable in 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

Deep Learning-Based Volatility Forecasting, Portfolio Management, and Reinforcement Learning-Based Risk Optimisation

  • 期刊论文

  • 发表日期：

    2026年1月3日

查看全文
  • Rahul Jain,
  • Nitin Varshney,
  • M. S. P. Durgarao,
  • Satish Kumar Maurya,
  • Deepak Kumar Mehta,
  • Aurghyadip Kundu,
  • Madhurima Halder,
  • Ankita Verma

摘要

In this paper, a hybrid method of equity market analysis with deep learning for volatility forecasting and reinforcement learning for portfolio optimization is presented. Volatility forecasting has been performed using Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models on past S&P 500 and VIX index data (2015–2024). GRU performs better than LSTM in terms of the forecast with an R2 of 0.91. These predictions are aggregated in a trading platform by a Proximal Policy Optimization (PPO) agent that learns to adjust trading strategies from dynamic market indicators like realized volatility, RSI, and MACD. The PPO agent performs outstandingly in portfolio appreciation, increasing capital from $10,000 to over $22,000 with a modest 28.9% winning rate. There is a huge risk-reward ratio and solid decision-making due to it. The research is adequately supported by the appropriateness of sequence models and reinforcement learning towards designing adaptive, intelligent trading algorithms that are highly generalizable in volatile markets.