Paper: arXiv 2408.06634

Authors: Haowei Ni, Shuchen Meng, Xupeng Chen, Ziqing Zhao, Andi Chen, Panfeng Li, Shiyao Zhang, Qifu Yin, Yuanqing Wang, Yuxi Chan

Abstract

Accurate stock market predictions following earnings reports are crucial for investors. Traditional methods, particularly classical machine learning models, struggle with these predictions because they cannot effectively process and interpret extensive textual data contained in earnings reports and often overlook nuances that influence market movements. This paper introduces an advanced approach by employing Large Language Models (LLMs) instruction fine-tuned with a novel combination of instruction-based techniques and quantized low-rank adaptation (QLoRA) compression. Our methodology integrates ‘base factors’, such as financial metric growth and earnings transcripts, with ’external factors’, including recent market indices performances and analyst grades, to create a rich, supervised dataset. This comprehensive dataset enables our models to achieve superior predictive performance in terms of accuracy, weighted F1, and Matthews correlation coefficient (MCC), especially evident in the comparison with benchmarks such as GPT-4. We specifically highlight the efficacy of the llama-3-8b-Instruct-4bit model, which showcases significant improvements over baseline models. The paper also discusses the potential of expanding the output capabilities to include a ‘Hold’ option and extending the prediction horizon, aiming to accommodate various investment styles and time frames. This study not only demonstrates the power of integrating cutting-edge AI with fine-tuned financial data but also paves the way for future research in enhancing AI-driven financial analysis tools.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper demonstrates high empirical rigor through detailed data collection, preprocessing, textualization, and comparative evaluation against benchmarks using standard metrics. However, its mathematical complexity is relatively low, focusing on implementing and adapting existing LLM techniques (QLoRA, instruction tuning) rather than deriving novel theoretical models.

Research Flowchart

  flowchart TD
  A["Research Goal: Accurate Stock Prediction<br>from Earnings Reports"] --> B["Data Collection & Integration"]
  B --> B1["Base Factors<br>Financial Metrics, Transcripts"]
  B --> B2["External Factors<br>Market Indices, Analyst Grades"]
  B1 & B2 --> C["Model Architecture<br>Llama-3-8B-Instruct-4bit"]
  C --> D["Fine-tuning Strategy<br>QLoRA + Instruction Tuning"]
  D --> E["Model Evaluation"]
  E --> F{"Outcomes & Findings"}
  F --> F1["Superior Metrics<br>Accuracy, F1, MCC"]
  F --> F2["SOTA Comparison<br>Outperforms GPT-4"]
  F --> F3["Future Directions<br>Hold Option, Horizon Extension"]

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 instruction‑fine‑tuned large language models compressed with QLoRA, using integrated base and external financial factors for earnings‑report stock prediction.

Source sentence
This paper introduces an advanced approach by employing Large Language Models (LLMs) instruction fine-tuned with a novel combination of instruction-based techniques and quantized low-rank adaptation (QLoRA) compression.

Assumes

No explicit assumptions found in the text.

Data

  • Base text-news unknown
    Source sentence
    Two distinct datasets, named Base and Full, were prepared to assess the impact of various features on model performance.
  • Full text-news unknown
    Source sentence
    Two distinct datasets, named Base and Full, were prepared to assess the impact of various features on model performance.

Costs

Not applicable The paper reports no trading or execution result.

Finds 1

  • The llama‑3‑8b‑Instruct‑4bit model achieved 0.573 accuracy, 0.565 weighted F1 and 0.154 MCC on the Full dataset, outperforming GPT‑4 by 16% higher accuracy and 10% better weighted F1. performance
    Source sentence
    It achieved 16% higher accuracy and 10% better Weighted F1 compared to GPT-4, indicating its effectiveness in handling complex, multifaceted financial data.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 4 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive (2 by independent authors)

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.

Cited References (13)

  1. LoRA: Low-Rank Adaptation of Large Language Models (2021) — link
  2. QLoRA: Efficient Finetuning of Quantized LLMs (2023) — link
  3. From Twitter to Reasoner: Understand Mobility Travel Modes and Sentiment Using Large Language Models (2024)
  4. Adaptive Ensembles of Fine-Tuned Transformers for LLM-Generated Text Detection (2024) — link
  5. Enhancing Financial Sentiment Analysis via Retrieval Augmented Large Language Models — → 2310.04027
  6. Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models — → 2402.03659
  7. Large language models can learn temporal reasoning (2024) — link
  8. Hadamard adapter: An extreme parameter-efficient adapter tuning method for pre-trained language models (2023)
and 5 more
  1. Predicting Stock Prices with FinBERT-LSTM: Integrating News Sentiment Analysis (2024) — link
  2. Fine-Tuning Gemma-7B for Enhanced Sentiment Analysis of Financial News Headlines (2024) — link
  3. Surveying Attitudinal Alignment Between Large Language Models Vs. Humans Towards 17 Sustainable Development Goals (2024) — link
  4. Twitter Sentiment analysis of covid vaccines (2021)
  5. Pruning as a Domain-specific LLM Extractor (2024) — link

2 of 13 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.