Paper: arXiv 2605.01384

Authors: Jinfeng Pan, Jiahao Chen

Abstract

Portfolio optimization is constrained by linear assumptions and insufficient integration of multi-modal information in traditional models. This paper proposes a cross-modal BERT-driven Actor-Critic framework SBCA for multi-asset portfolio optimization to address the deficiencies of existing deep reinforcement learning DRL methods in fusing price data and financial text sentiment, as well as lacking practical trading constraints. The framework adopts a cross-modal gated fusion mechanism to adaptively integrate price time-series features and text semantic features, embeds downside risk and turnover penalty constraints into the reward function, and constructs a complete empirical system for validation. Experiments on 11-year U.S. stock multi-asset datasets show that SBCA outperforms equal weight, buy-and-hold and market benchmark strategies in portfolio value, annual return, Sharpe ratio and maximum drawdown. Ablation studies verify the complementary enhancement of Actor-Critic mechanism and cross-modal fusion module. Cost sensitivity analysis confirms the model’s robustness under varying transaction costs. SBCA provides an effective and interpretable end-to-end solution for dynamic quantitative portfolio decision-making.

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The research combines sophisticated neural architectures (Actor-Critic DRL and gated cross-modal BERT fusion) with extensive empirical testing, including cost sensitivity analysis, ablation studies, and 11-year multi-asset backtesting. The blend of advanced machine learning theory and rigorous financial performance validation places this paper firmly in a high-math, high-rigor classification.

Research Flowchart

  flowchart TD
    A[<b>Research Goal:</b> Improve Multi-Asset <br/>Portfolio Optimization via SBCA] --> B[<b>Data Inputs:</b> Price Time-Series & <br/>Financial Text Sentiment]
    B --> C[<b>Methodology:</b> Cross-modal Gated <br/>Fusion & DRL Actor-Critic]
    C --> D[<b>Processes:</b> Incorporate Downside Risk <br/>& Turnover Penalties]
    D --> E[<b>Validation:</b> 11-Year Empirical Study <br/>& Ablation Analysis]
    E --> F[<b>Outcomes:</b> Superior Risk-Adjusted Returns <br/>& High Model Robustness]

Cited References (36)

  1. Deep portfolio management: A deep reinforcement learning framework for the financial portfolio management problem (2017) — link
  2. Model-based deep reinforcement learning for dynamic portfolio optimization (2019) — link
  3. Cost-sensitive portfolio selection via deep reinforcement learning (2020)
  4. Dynamic portfolio rebalancing through reinforcement learning (2022)
  5. A reinforcement learning approach to dynamic portfolio optimization (2025)
  6. Portfolio management using deep reinforcement learning — → 2405.01604
  7. Reinforcement learning-based cryptocurrency portfolio management using soft actor–critic and deep deterministic policy gradient algorithms — → 2511.20678
  8. Risk-adjusted deep reinforcement learning for portfolio optimization: A multi-reward approach (2025)
and 28 more
  1. Deep reinforcement learning for optimal portfolio allocation: A comparative study with mean-variance optimization — → 2602.17098
  2. Finrl: Deep reinforcement learning framework to automate trading in quantitative finance (2021)
  3. FinRL: A deep reinforcement learning library for automated stock trading in quantitative finance (2022) — link
  4. FinRL-meta: Market environments and benchmarks for data-driven financial reinforcement learning (2022)
  5. Finbert: Financial sentiment analysis with pre-trained language models (2019) — link
  6. Finbert: A pretrained language model for financial communications (2020) — link
  7. Finbert: A pre-trained financial language representation model for financial text mining (2020)
  8. Bert: Pre-training of deep bidirectional transformers for language understanding (2019)
  9. Stock movement prediction with financial news using contextualized embedding from BERT (2021) — link
  10. Stock trend prediction using financial market news and BERT (2020)
  11. Applying BERT to analyze investor sentiment in stock market (2020)
  12. Financial sentiment analysis on news and reports using large language models and FinBERT (2024) — link
  13. Enhancing sentiment analysis in stock market tweets through BERT-based knowledge transfer (2026)
  14. Multimodal deep reinforcement learning for portfolio optimization — → 2412.17293
  15. Cross-modal temporal fusion for financial market forecasting — → 2504.13522
  16. Modality-aware transformer for financial time series forecasting (2024)
  17. Sentiment-aware portfolio optimization: CVaR-based diversification with deep reinforcement learning (2025)
  18. A multimodal deep fusion method for stock movement prediction using heterogeneous data source (2026)
  19. FLAG-TRADER: Fusion LLM-agent with gradient-based reinforcement learning for financial trading (2025)
  20. Portfolio selection (1952)
  21. Optimal versus naive diversification: How inefficient is the 1/N portfolio strategy? (2009)
  22. Portfolio selection with transactions costs (1976)
  23. Markowitz’s “Portfolio Selection”: A fifty-year retrospective (2002)
  24. Dynamic Programming and Statistical Communication Theory (1957)
  25. Reinforcement Learning: An Introduction (1998)
  26. Deep learning (2016)
  27. Deeptrader: A deep reinforcement learning approach for risk-return balanced portfolio management with market conditions embedding (2021)
  28. Handbook of mathematical functions with formulas, graphs, and mathematical tables (1948)

5 of 36 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.