Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FinLLM-B: When Large Language Models Meet Financial Breakout Trading

Kang Zhang, Osamu Yoshie, Lichao Sun, Weiran Huang

TL;DR

This work tackles the challenge of reliable financial breakout detection, where true breakouts must be distinguished from false signals with transparent reasoning. It introduces FinLLM-B, the first financial breakout–oriented large language model, and a novel multi-stage structure that splits reasoning and report generation to reduce errors and improve stability. A dedicated FinLLM-B dataset and a separate Report Generator dataset are constructed from footprint-chart data, enabling domain-aware reasoning and explainable final reports. Empirical results show substantial gains over GPT-3.5 and GPT-4, with average accuracy gains around 50% and a notable contribution from the multi-stage design to both accuracy and stability, indicating strong potential for practical, explainable breakout trading systems.

Abstract

Trading range breakout is a key method in the technical analysis of financial trading, widely employed by traders in financial markets such as stocks, futures, and foreign exchange. However, distinguishing between true and false breakout and providing the correct rationale cause significant challenges to investors. Traditional quantitative methods require large amounts of data and cannot directly present the reasoning process, making them less than perfect in this field. Recently, large language models have achieved success in various downstream applications, but their effectiveness in the domain of financial breakout detection has been subpar. The reason is that the unique data and specific knowledge are required in breakout detection. To address these issues, we create the first financial breakout dataset and introduce FinLLM-B, the premier large language model for financial breakout detection, which enhances the effectiveness of breakout trading strategies. Furthermore, we have developed a novel framework for large language models, namely multi-stage structure, effectively reducing mistakes in downstream applications. Experimental results indicate that compared to GPT-3.5, FinLLM-B improves the average accuracy of answers and rational by 49.97%, with the multi-stage structure contributing 9.72% to the improvement. Additionally, it outperforms ChatGPT-4 by 42.38%.

FinLLM-B: When Large Language Models Meet Financial Breakout Trading

TL;DR

This work tackles the challenge of reliable financial breakout detection, where true breakouts must be distinguished from false signals with transparent reasoning. It introduces FinLLM-B, the first financial breakout–oriented large language model, and a novel multi-stage structure that splits reasoning and report generation to reduce errors and improve stability. A dedicated FinLLM-B dataset and a separate Report Generator dataset are constructed from footprint-chart data, enabling domain-aware reasoning and explainable final reports. Empirical results show substantial gains over GPT-3.5 and GPT-4, with average accuracy gains around 50% and a notable contribution from the multi-stage design to both accuracy and stability, indicating strong potential for practical, explainable breakout trading systems.

Abstract

Trading range breakout is a key method in the technical analysis of financial trading, widely employed by traders in financial markets such as stocks, futures, and foreign exchange. However, distinguishing between true and false breakout and providing the correct rationale cause significant challenges to investors. Traditional quantitative methods require large amounts of data and cannot directly present the reasoning process, making them less than perfect in this field. Recently, large language models have achieved success in various downstream applications, but their effectiveness in the domain of financial breakout detection has been subpar. The reason is that the unique data and specific knowledge are required in breakout detection. To address these issues, we create the first financial breakout dataset and introduce FinLLM-B, the premier large language model for financial breakout detection, which enhances the effectiveness of breakout trading strategies. Furthermore, we have developed a novel framework for large language models, namely multi-stage structure, effectively reducing mistakes in downstream applications. Experimental results indicate that compared to GPT-3.5, FinLLM-B improves the average accuracy of answers and rational by 49.97%, with the multi-stage structure contributing 9.72% to the improvement. Additionally, it outperforms ChatGPT-4 by 42.38%.
Paper Structure (26 sections, 6 figures, 3 tables)

This paper contains 26 sections, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Trading Range Breakout.
  4. Large Language Models.
  5. Problem Formulation
  6. Method
  7. Multi-Stage Structure
  8. Task Flow.
  9. Report Generator.
  10. Dataset
  11. FinLLM-B Dataset.
  12. Report Generator Dataset.
  13. Experiment
  14. Baseline & Evaluation Metrics.
  15. Professionalism evaluation.
  16. ...and 11 more sections

Figures (6)

  • Figure 1: A sample of FinLLM-B task.
  • Figure 2: Overview of FinLLM-B with multi-stage structure. Multi-stage structure consists of two parts: Part A and Part B. Part A comprises three stages, each corresponding to a subtask of breakout detection. Part B is responsible for integrating the answers from Part A into a rationale and providing the final answer.
  • Figure 3: Dataset Construction. A: Footprint chart. B: Data table derived from the footprint chart. C: Dataset. It consists of two parts: FinLLM-B dataset and report generator dataset.
  • Figure 4: Output samples for professionalism comparison. Green: Valuable domain knowledge. Red: Incorrect domain knowledge and mistakes. Due to the length of the output, we used '...' to omit non-essential content.
  • Figure 5: Accuracy comparison. Each axis is rescaled independently for better comparison.
  • ...and 1 more figures