Beyond Polarity: Multi-Dimensional LLM Sentiment Signals for WTI Crude Oil Futures Return Prediction

2026-03-12 · Source: Takara TLDR - Daily AI Papers · Field: Finance & Economics — Commodities & Energy Finance, Capital Markets & Investment Management, FinTech & Digital Financial Services · Depth: Advanced, quick

Summary

A recent study investigates whether multi-dimensional sentiment signals, extracted by large language models, can enhance the prediction of weekly WTI crude oil futures returns. Researchers analyzed energy-sector news articles published between 2020 and 2025, using GPT-4o, Llama 3.2-3b, FinBERT, and AlphaVantage to construct five sentiment dimensions: relevance, polarity, intensity, uncertainty, and forwardness. These article-level signals were aggregated weekly and evaluated within a classification framework. The most effective predictive performance was achieved by combining GPT-4o and FinBERT, indicating that LLM-based and traditional financial sentiment models offer complementary information. SHAP analysis highlighted intensity- and uncertainty-related features as key predictors, demonstrating that news sentiment's predictive value extends beyond basic polarity.

Key takeaway

For an AI Scientist developing commodity price prediction models, you should integrate multi-dimensional sentiment analysis from large language models. Specifically, consider combining outputs from models like GPT-4o with traditional financial sentiment models such as FinBERT to capture complementary predictive signals. Prioritize features related to sentiment intensity and uncertainty, as these have shown significant predictive power beyond simple positive or negative polarity in WTI crude oil futures.

Key insights

Multi-dimensional LLM sentiment, beyond polarity, improves WTI crude oil futures return prediction.

Principles

• Sentiment dimensions are complementary.
• Intensity and uncertainty are key predictors.

Method

Five sentiment dimensions (relevance, polarity, intensity, uncertainty, forwardness) were extracted from news using LLMs and benchmark models, then aggregated weekly for classification.

In practice

• Combine GPT-4o and FinBERT for forecasting.
• Focus on intensity and uncertainty signals.

Topics

• Large Language Models
• Sentiment Analysis
• Financial Forecasting
• Crude Oil Futures
• WTI Crude Oil Prediction

Best for: NLP Engineer, AI Scientist, Research Scientist, AI Researcher, Data Scientist, AI Data Scientist

Related on AIssential

• Enhancing Brazilian Inflation Forecasts through Sentiment Analysis Using Large Language Models — LLM-derived sentiment from policy minutes significantly enhances inflation forecasting accuracy when aligned with human assessments.
• Semantics-Enhanced Retrieval-Augmented Time Series Forecasting — Time series forecasting improves by combining numerical similarity with semantic context from textual descriptions.
• Speaking Numbers to LLMs: Multi-Wavelet Number Embeddings for Time Series Forecasting — TempoWave improves LLM time series forecasting by aligning continuous numerical data with discrete language models using multi-wavelet embeddings.
• Open Thread 424 — Global energy security is robust, with multiple mechanisms mitigating severe oil supply shocks from geopolitical events.
• Data Machina #261 — Generative AI and LLMs are being adapted to achieve state-of-the-art performance in diverse time-series tasks.
• What sentiment analysis can't see: Measuring whether customers were helped, and what went wrong, across 70,000 support conversations — LLM-based analysis of support conversations surpasses sentiment analysis in measuring customer satisfaction and identifying underlying problems.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.