Academic Alpha

Can ChatGPT Predict Commodity Returns? Yes.

This paper introduces a ChatGPT-based Commodity News Ratio Index (CNRI) to forecast excess returns on commodity futures. Using over 2.5 million articles across 18 commodities and 80 years, the CNRI significantly outperforms traditional predictors both in- and out-of-sample.

💡 Takeaway:
ChatGPT captures commodity sentiment better than BERT, BoW, or macroeconomic variables—offering robust, predictive signals for commodity futures returns.

Key Idea: What Is This Paper About?

The paper constructs a Commodity News Ratio Index (CNRI) by analyzing 2.5M news articles using ChatGPT. Each article is scored based on whether ChatGPT believes it signals a price increase or decrease. These sentiment signals are aggregated into a commodity-level CNRI and distilled using Partial Least Squares (PLS) to forecast commodity futures index excess returns.

Economic Rationale: Why Should This Work?

📌 Relevant Economic Theories and Justifications:

  • Investor Sentiment: Media sentiment influences commodity markets, especially when driven by global narratives.
  • LLMs Extract Informational Signals: ChatGPT outperforms classical NLP (BoW, BERT) by capturing deeper semantics and context.
  • Slow News Diffusion: Commodities are less covered in financial media, allowing sentiment to diffuse slowly into prices.
  • Predictive Link to Macro: CNRI also predicts GDP, IPI, CFNAI, and PPI—suggesting macroeconomic underpinnings.

📌 Why It Matters:
This approach provides forward-looking, interpretable, and systematic sentiment-driven signals using LLMs—extending NLP-based forecasting to commodities.

How to Do It: Data, Model, and Strategy Implementation

Data Used

  • Text Corpus: 2.6M articles (1946–2022) from 9 major newspapers
  • Commodities: 18 (e.g., crude oil, gold, soybeans, zinc)
  • Returns: Monthly commodity futures index excess returns (Levine et al. 2018)

Model / Methodology

  • Step 1: Use ChatGPT to classify articles as “GOING UP”, “GOING DOWN”, or “UNKNOWN”
  • Step 2: Compute a Commodity News Ratio (CNR) = (Good − Bad) / Total articles (past 3 months)
  • Step 3: Use PLS to form the ChatGPT-based CNRI from 18 CNRs
  • Step 4: Forecast commodity index returns using CNRI
  • Comparison Models: BERT-based CNRI, BoW-based CNRI, macroeconomic indicators, business cycle controls

Trading Strategy (Investor Application)

  • Signal: Use CNRI to forecast next 1–12 months of commodity index returns
  • Portfolio Construction:
    • Allocate between commodity futures and T-bills
    • Use CNRI forecasts in mean-variance optimization
    • Volatility estimated using 12-month rolling GSCI standard deviation
  • Performance:
    • Highest Sharpe: 0.407 (3-month horizon)
    • Highest CER: 5.86% (1-month horizon)
    • Robust across macro states (recessions, backwardation, inflation down)

Key Table or Figure from the Paper

📊 Reference: [Table 7] – Out-of-Sample Forecasting and Portfolio Performance

📌 Explanation:

  • Shows the CNRI achieves out-of-sample R² from 2.09% to 5.84%
  • Sharpe Ratios exceed benchmark across all horizons
  • CER gains peak at 2.1%–2.2% over 6–9 month horizons
  • CNRI outperforms both economic indicators and BERT-based sentiment

Final Thought

💡 ChatGPT decodes commodity headlines into portfolio alpha—NLP for futures trading is here. 🚀

Paper Details (For Further Reading)

  • Title: ChatGPT and Commodity Return
  • Authors: Shen Gao, Shijie Wang, Yuanzhi Wang, Qunzi Zhang
  • Publication Year: 2025
  • Journal/Source: Journal of Futures Markets
  • Link: https://doi.org/10.1002/fut.22568
Share Tweet Share Share Email

Read next