Not All News Is Equal: Topic- and Event-Conditional Sentiment from Finetuned LLMs for Aluminum Price Forecasting
“Our results demonstrate that during periods of high volatility, Long Short-Term Memory (LSTM) models incorporating sentiment data from a finetuned Qwen3 model (Sharpe ratio 1.04) significantly outperform baseline models using tabular data alone (Sharpe ratio 0.23).”
AssetCurrenciesindices, currency pairs
StrategyMachine learningnews sentiment
TechniqueLarge language modelrecurrent neural network
GeographyChina
Also stated: Monthly, 2007–2024, Sharpe ratio, Excess return, Long-short, Macro, Text and news
Missing from abstract: benchmark, out-of-sample test, costs, multiple testing, universe
Institutional-grade audit of this strategy.
£5 — Request the audit- Audit of the paper claims accounting for costs, luck, crises, originality, geography, etc.
- Reproduction code integratable into Bloomberg, IBKR, etc.
- Bias-free datasets for backtesting included
- Delivered in 2 working days
The paper's neighbourhoodMachine learning return prediction by year
The family is counted across the whole universe. Asset, method and authors are counted inside the family.
| Similar papers | citation | same method | same asset | same authors | |
|---|---|---|---|---|---|
| Spatiotemporal Signals in Residuals: A Two-Stage Boosting Graph Neural Network for Stock Return Prediction by Liu et al. | 2026 | ||||
| Machine Learning for Cross-Sectional Return Prediction: A Comparative Study of US and Indian Equities by Khatu | 2026 | ||||
| Empirical Asset Pricing via Machine Learning: Evidence from Indian Equity Markets by Sharma | 2026 | ||||
| Information Diffusion and Return Predictability across Business and Ownership Networks: Evidence from Korea by Lee and Lim | 2026 | ||||
| DistillSem: LLM-Distilled Financial Headline Semantics for Cross-Sectional Return Prediction by Guo | 2026 |