Scholarlink

<p>Financial markets exhibit regime shifts characterized by low, medium, and high volatility, limiting the effectiveness of static portfolio strategies. This study evaluates five deep reinforcement learning (DRL) algorithms—A2C, PPO, DDPG, TD3, and SAC—applied to NIFTY50 constituents within a trading environment incorporating transaction costs, position limits, and turbulence-based risk penalties. Portfolio performance is examined across volatility regimes identified using rolling variance and DCC-GARCH conditional variance, and benchmarked against both the NIFTY50 index and a dynamic minimum-variance portfolio. The results show that DRL agents outperform the market benchmark during most medium- and low-volatility regimes, but their relative performance varies systematically across market conditions, with no single algorithm consistently dominant. During periods of heightened volatility, variance-based portfolios provide stronger downside protection. Overall, the findings highlight the importance of regime-aware and adaptive portfolio frameworks that combine learning-based strategies with risk-focused approaches.</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

Volatility Regime-Dependent Portfolio Optimization: Empirical Analysis of Deep Reinforcement Learning Agents

  • 期刊论文

  • 发表日期：

    2026年7月27日

查看全文
  • Iqbal K. I.,
  • Remya Ramachandran

摘要

Financial markets exhibit regime shifts characterized by low, medium, and high volatility, limiting the effectiveness of static portfolio strategies. This study evaluates five deep reinforcement learning (DRL) algorithms—A2C, PPO, DDPG, TD3, and SAC—applied to NIFTY50 constituents within a trading environment incorporating transaction costs, position limits, and turbulence-based risk penalties. Portfolio performance is examined across volatility regimes identified using rolling variance and DCC-GARCH conditional variance, and benchmarked against both the NIFTY50 index and a dynamic minimum-variance portfolio. The results show that DRL agents outperform the market benchmark during most medium- and low-volatility regimes, but their relative performance varies systematically across market conditions, with no single algorithm consistently dominant. During periods of heightened volatility, variance-based portfolios provide stronger downside protection. Overall, the findings highlight the importance of regime-aware and adaptive portfolio frameworks that combine learning-based strategies with risk-focused approaches.