Paper: arXiv 2603.11408

Authors: Dehao Dai, Ding Ma, Dou Liu, Kerui Geng, Yiqing Wang

Abstract

Forecasting crude oil prices remains challenging because market-relevant information is embedded in large volumes of unstructured news and is not fully captured by traditional polarity-based sentiment measures. This paper examines whether multi-dimensional sentiment signals extracted by large language models improve the prediction of weekly WTI crude oil futures returns. Using energy-sector news articles from 2020 to 2025, we construct five sentiment dimensions covering relevance, polarity, intensity, uncertainty, and forwardness based on GPT-4o, Llama 3.2-3b, and two benchmark models, FinBERT and AlphaVantage. We aggregate article-level signals to the weekly level and evaluate their predictive performance in a classification framework. The best results are achieved by combining GPT-4o and FinBERT, suggesting that LLM-based and conventional financial sentiment models provide complementary predictive information. SHAP analysis further shows that intensity- and uncertainty-related features are among the most important predictors, indicating that the predictive value of news sentiment extends beyond simple polarity. Overall, the results suggest that multi-dimensional LLM-based sentiment measures can improve commodity return forecasting and support energy-market risk monitoring.

Complexity vs Empirical Score

  • Math Complexity: 2.5/10
  • Empirical Rigor: 6.5/10
  • Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper applies modern LLM techniques (GPT-4o, Llama) to extract multi-dimensional sentiment (uncertainty, forwardness) beyond basic polarity for crude oil futures prediction. While mathematical formulation is minimal, the empirical design uses clear stratification, model comparison, and SHAP explainability analysis across a multi-year dataset.

Research Flowchart

  flowchart TD
    A[Research Goal: Predict WTI Crude Oil Returns via Multi-Dimensional LLM Sentiment] --> B[Data Input: Energy News Articles 2020-2025]
    B --> C[Sentiment Extraction: Relevance, Polarity, Intensity, Uncertainty, Forwardness]
    C --> D[Models Used: GPT-4o, Llama 3.2-3b, FinBERT, AlphaVantage]
    D --> E[Aggregation & Classification: Weekly Signals & Return Forecasting]
    E --> F[Feature Analysis: SHAP Interpretability]
    F --> G[Outcomes: GPT-4o + FinBERT Best Hybrid; Intensity & Uncertainty Outperform Polarity]

Cited References (98)

  1. A unified approach to interpreting model predictions (2017)
  2. Why do tree-based models still outperform deep learning on typical tabular data? (2022)
  3. Tabular data: Deep learning is not all you need (2022)
  4. Investor sentiment and the cross-section of stock returns (2006)
  5. What’s in the news? Using news sentiment momentum for tactical asset allocation (2015)
  6. The llama 3 herd of models (2024) — link
  7. FinBERT: A large language model for extracting information from financial text (2023)
  8. Gpt-4 technical report (2023) — link
and 90 more
  1. Alpha Vantage API: Financial Market Data and News Sentiment (2026)
  2. Impact of sentiment analysis on energy sector stock prices: a FinBERT approach (2024)
  3. NTUSD-Fin: a market sentiment dictionary for financial social media data applications (2018)
  4. Which risk factors drive oil futures price curves? (2020) — link
  5. Text-based crude oil price forecasting: A deep learning approach (2019) — link
  6. Oil prices, economic activity and inflation: evidence for some Asian countries (2005) — link
  7. Oil prices, economic activity and inflation: evidence for some Asian countries (2005) — link
  8. Jumps and oil futures volatility forecasting: a new insight (2020)
  9. Forecasting crude oil futures market returns: A principal component analysis combination approach (2022) — link
  10. The role of China ’s crude oil futures in world oil futures market and China ’s financial market (2023) — link
  11. Prediction of Crude Oil Price using LLM : An Empirical Analysis (2025) — link
  12. Giving Content to Investor Sentiment : The Role of Media in the Stock Market (2007) — link
  13. Cooperative Sentiment Agents for Multimodal Sentiment Analysis (2024) — link
  14. The relationship between individual investor sentiment, stock return and volatility: Evidence from the Turkish market (2015) — link
  15. Comparing LLMs for Sentiment Analysis in Financial Market News — → 2510.15929
  16. Twitter mood predicts the stock market (2011) — link
  17. Sentiment analysis of Twitter data for predicting stock market movements (2016) — link
  18. Generative Agents : Interactive Simulacra of Human Behavior (2023) — link
  19. U. S . banks’ lending, financial stability, and text-based sentiment analysis (2022) — link
  20. Sentiment Analysis of Online Lending Services Using Support Vector Machine and Logistic Regression (2025) — link
  21. Lender Trust on the P2P Lending : Analysis Based on Sentiment Analysis of Comment Text (2020) — link
  22. A three-stage prediction model for firm default risk: An integration of text sentiment analysis (2025) — link
  23. Forecasting crude oil prices: A reduced-rank approach (2023)
  24. Oil and the macroeconomy since World War II (1983)
  25. Oil and the macroeconomy since the 1970s (2002)
  26. Application of ARIMA model in short-term prediction of international crude oil price (2013)
  27. Volatility in crude oil futures: A comparison of the predictive ability of GARCH and implied volatility models (2009)
  28. A deep learning ensemble approach for crude oil price forecasting (2017)
  29. Forecasting crude oil futures price using machine learning methods: Evidence from China (2023)
  30. Effective crude oil price forecasting using new text-based and big-data-driven model (2021)
  31. Energy organization sentiment and oil return forecast (2025)
  32. A novel secondary decomposition method for forecasting crude oil price with twitter sentiment (2024)
  33. LLMs analyzing the analysts: Do BERT and GPT extract more value from financial analyst reports? (2023)
  34. A novel forecasting framework leveraging large language model and machine learning for methanol price (2025)
  35. Sentiment analysis for stock price prediction (2020)
  36. Social media sentiment, investor herding and informational efficiency (2026)
  37. Sentiment analysis on social media for stock movement prediction (2015)
  38. The association between stock price volatility and financial news–a sentiment analysis approach (2017)
  39. FinSenticNet: A concept-level lexicon for financial sentiment analysis (2023)
  40. Optuna: A next-generation hyperparameter optimization framework (2019)
  41. Algorithms for hyper-parameter optimization (2011)
  42. Lightgbm: A highly efficient gradient boosting decision tree (2017)
  43. When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks (2011)
  44. Measuring economic policy uncertainty (2016)
  45. Llama: Open and efficient foundation language models (2023) — link
  46. Can chatgpt forecast stock price movements? return predictability and large language models — → 2304.07619
  47. Forecasting the term structure of crude oil futures prices with neural networks (2016)
  48. Research on crude oil futures price prediction methods: A perspective based on quantum deep learning (2025)
  49. A new approach for forecasting crude oil prices based on stochastic and deterministic influences of LMD using ARIMA and LSTM models (2023)
  50. A novel dynamic time-delay grey model of energy prices and its application in crude oil price forecasting (2022)
  51. Multi-step-ahead crude oil price forecasting using a hybrid grey wave model (2018)
  52. A new hybrid deep learning model for monthly oil prices forecasting (2023)
  53. News media and attention spillover across energy markets: A powerful predictor of crude oil futures prices (2022)
  54. Text-based crude oil price forecasting: A deep learning approach (2019)
  55. Forecasting crude oil futures prices using global macroeconomic news sentiment (2020)
  56. Textual sentiment in finance: A survey of methods and models (2014)
  57. Textual analysis and machine leaning: Crack unstructured data in finance and accounting (2016)
  58. Sentiment analysis and machine learning in finance: a comparison of methods and models on one million messages (2020)
  59. Stock returns and investor sentiment: textual analysis and social media (2020)
  60. Intraday online investor sentiment and return patterns in the US stock market (2017)
  61. Textual analysis in accounting and finance: A survey (2016)
  62. Sentiment analysis algorithms and applications: A survey (2014)
  63. More than words: Quantifying language to measure firms’ fundamentals (2008)
  64. Reading between the lines: An empirical examination of qualitative attributes of financial analysts’ reports (2012)
  65. Divergence of sentiment and stock market trading (2017)
  66. Is all that talk just noise? The information content of internet stock message boards (2004)
  67. The information content of forward-looking statements in corporate filings—A na "\i ve Bayesian machine learning approach (2010)
  68. Tweets and trades: The information content of stock microblogs (2014)
  69. Stock return predictability and investor sentiment: A high-frequency perspective (2016)
  70. Sentiment spillover effects for US and European companies (2019)
  71. When machines read the news: Using automated text analytics to quantify high frequency news-implied market reactions (2011)
  72. How to fine-tune bert for text classification? (2019)
  73. BERT-based financial sentiment index and LSTM-based stock return predictability (2019) — link
  74. Text-based sentiment analysis in finance: Synthesising the existing literature and exploring future directions (2024)
  75. Multidimensional Investor Sentiment and Stock Market Volatility in China: Measurement, Mechanisms and Applications (2025)
  76. Sentimental Analysis to Predict Stock Market Using in Neutrosophic Time Series. (2025)
  77. Data-Driven Insights: Leveraging Sentiment Analysis and Latent Profile Analysis for Financial Market Forecasting (2026)
  78. Deep learning-based multi-dimensional investor sentiment and stock liquidity: Evidence from China (2025)
  79. An intertemporal capital asset pricing model (1973)
  80. Uncertainty and risk in financial markets (2005)
  81. Information uncertainty and expected returns (2005)
  82. Information uncertainty and analyst forecast behavior (2006)
  83. Fundamental uncertainty and stock market volatility (2008)
  84. Treasury bond volatility and uncertainty about monetary policy (2010)
  85. Economic policy uncertainty and stock market liquidity: Evidence from G7 countries (2021)
  86. The effect of uncertainty on stock market volatility and correlation (2023)
  87. Dual effects of investor sentiment and uncertainty in financial markets (2024)
  88. Sentiment and uncertainty (2022)
  89. Good and bad uncertainty: Macroeconomic and financial market implications (2015)
  90. News implied volatility and disaster concerns (2017)

2 of 98 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.