Paper: arXiv 2508.07408
Authors: Yueyi Wang, Qiyao Wei
Abstract
In this study, we wish to showcase the unique utility of large language models (LLMs) in financial semantic annotation and alpha signal discovery. Leveraging a corpus of company-related tweets, we use an LLM to automatically assign multi-label event categories to high-sentiment-intensity tweets. We align these labeled sentiment signals with forward returns over 1-to-7-day horizons to evaluate their statistical efficacy and market tradability. Our experiments reveal that certain event labels consistently yield negative alpha, with Sharpe ratios as low as -0.38 and information coefficients exceeding 0.05, all statistically significant at the 95% confidence level. This study establishes the feasibility of transforming unstructured social media text into structured, multi-label event variables. A key contribution of this work is its commitment to transparency and reproducibility; all code and methodologies are made publicly available. Our results provide compelling evidence that social media sentiment is a valuable, albeit noisy, signal in financial forecasting and underscore the potential of open-source frameworks to democratize algorithmic trading research.
Complexity vs Empirical Score
- Math Complexity: 3.5/10
- Empirical Rigor: 6.0/10 (capped from 7.2: transaction costs not modelled; no out-of-sample test; reported Sharpe ratio of 5 is implausibly high for daily or lower-frequency data; see penalty flags)
- Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 7.2)
Why this score: The paper focuses on practical implementation of LLM-based feature engineering with extensive backtesting metrics (Sharpe ratios, information coefficients) and transparency, but its mathematical depth is moderate, relying primarily on statistical testing rather than advanced derivations.
Research Flowchart
flowchart TD
A["Research Goal:<br>Extract Event-Aware Sentiment<br>Alpha from Financial Tweets"] --> B["Input: Company-Related<br>Twitter Data Corpus"]
B --> C["LLM-Augmented<br>Semantic Annotation<br>Assign Multi-Label Events"]
C --> D["Signal Alignment:<br>Link Sentiment to<br>1-7 Day Forward Returns"]
D --> E{"Statistical &<br>Tradability Evaluation"}
E -->|"Quantitative Metrics"| F["Key Findings/Outcomes"]
F --> G["Identified Negative Alpha<br>Event Labels (Sharpe ~ -0.38)"]
F --> H["Transparent Framework<br>Code & Methods Publicly Available"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a modular pipeline that uses large language models to assign multi‑label event categories to high‑sentiment tweets and builds interpretable, event‑driven sentiment factors for quant trading.
Source sentence
we use an LLM to automatically assign multi-label event categories to high-sentiment-intensity tweets.
Assumes
No explicit assumptions found in the text.
Data
US equities · equities · daily
- Tweet Corpus alternative public
Source sentence
We utilize the dataset from Sowinska et al. (Sowinska & Madhyastha, 2020), which includes 862,231 English-language tweets linked to stock tickers over a multi-year period. We restrict our study to a cleaned sub‑set of 85,176 tweets for higher signal-to-noise ratio.
- Market Data market-prices public
Source sentence
Aligned stock-level price and volume data are sourced from the same dataset. Daily logarithmic returns are computed as rt = log(Pt/Pt−1) where Pt is the closing price.
Twitter / X
Costs
Not modeled
Finds 4
Sharpe 5.00 · annual return 8.0% · max drawdown 15.2%
- Certain event labels consistently yield negative alpha, with Sharpe ratios as low as -0.38 and information coefficients exceeding 0.05, all statistically significant at the 95% confidence level. performance
Source sentence
Our experiments reveal that certain event labels consistently yield negative alpha, with Sharpe ratios as low as -0.38 and information coefficients exceeding 0.05, all statistically significant at the 95% confidence level.
- Speculation/Rumor event has a 1‑day Sharpe ratio of -0.337*** and IC of -0.039. performance
Source sentence
1-Day Speculation/Rumor 130 -0.337*** -0.039
- Retail Investor Buzz event has a 1‑day Sharpe ratio of -0.359*** and IC of 0.096. performance
Source sentence
1-Day Retail Investor Buzz 92 -0.359*** 0.096
- Geopolitical Tension event has a 1‑day Sharpe ratio of -0.661** and IC of -0.139. performance
Source sentence
1-Day Geopolitical Tension 25 -0.661** -0.139
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 16 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.