Can large language models (LLMs) generate enhanced returns by analyzing earnings releases? This thesis explores the intersection of AI-powered language models and earnings trading strategies. In a structured experiment, earnings releases from five American blue-chip stocks are evaluated by a selected LLM, classifying sentiment as positive, neutral, or negative, across 18 quarters from Q1 2019 to Q2 2023.
The theoretical foundation covers earnings trading strategies, the characteristics of American blue-chip stocks, and the core concepts of artificial intelligence, machine learning, deep learning, and LLMs. Three LLMs are assessed for suitability, and the most fitting model is selected for the experiment.
The LLM's sentiment outputs are then compared against actual stock price movements to determine whether AI-driven earnings analysis can meaningfully improve investment decisions. The findings are relevant to both retail traders seeking a competitive edge and institutional investors exploring scalable, AI-based approaches to market analysis.
Table of Contents
1 INTRODUCTION
1.1 Introduction to the Topic
1.2 Delimitation of the Thesis
2 EARNINGS TRADING STRATEGIES
2.1 Overview of Earnings Trading
2.1.1 Pre-Earnings Strategies
2.1.2 Post-Earnings Strategies
2.1.3 Conclusion about Earnings Trading
3 AMERICAN BLUE-CHIP STOCKS
3.1 Introduction to Blue-Chip Stocks
3.2 Benefits of American Blue-Chip Stocks
4 BASICS OF MACHINE LEARNING AND LARGE LANGUAGE MODELS
4.1 Fundamentals of Artificial Intelligence
4.1.1 Artificial Intelligence
4.1.2 Machine Learning
4.1.3 Deep Learning
4.1.4 Large Language Models
4.1.5 Importance of these Terms
4.2 Fundamentals of Machine Learning and LLMs
4.2.1 Supervised Learning
4.2.2 Unsupervised Learning
4.2.3 Reinforcement Learning
4.2.4 Neural Networks
4.3 Core Concepts of LLMs
5 EVALUATING THREE LLMS FOR SUITABILITY FOR EARNINGS TRADING
5.1 Overview of different LLMs
5.1.1 Bidirectional Encoder Representations
5.1.2 A Robustly Optimized BERT Approach
5.1.3 Generative Pre-trained Transformer
5.2 Evaluation of the three LLMs
6 ASSESSING EARNINGS RELEASES WITH THE CHOSEN LARGE LANGUAGE MODEL
6.1 Methodology to use an LLM to analyze Earnings Releases
6.1.1 Selection of five American Blue-Chip Stocks
6.1.2 Wording a Prompt to analyze Earnings Releases
6.1.3 Analyzing the Earnings Reports
6.1.4 Analyzing the Output of ChatGPT
7 ANALYSIS OF CHATGPT'S SENTIMENT RESPONSES
7.1 Assessment of ChatGPT's Sentiment Results
7.1.1 Amazon.com
7.1.2 The Coca-Cola Company
7.1.3 Walmart
7.1.4 Goldman Sachs
7.1.5 The Walt Disney Company
7.1.6 Overview of all stocks
7.2 Comparing the Sentiment Results with the Stock Movement post-announcement
7.3 Critical Evaluation of the Results
8 FUTURE RESEARCH
9 CONCLUSION ABOUT THE LLM-BASED EARNINGS TRADING EXPERIMENT
Objectives and Thematic Focus
This thesis explores the integration of Large Language Models (LLMs) into financial trading strategies, specifically by evaluating whether ChatGPT can analyze earnings reports to predict stock movements and provide actionable investment recommendations.
- The role of earnings releases in financial markets.
- Fundamental principles of Machine Learning and LLM architectures.
- Comparative evaluation of BERT, RoBERTa, and GPT models for financial sentiment analysis.
- Implementation of a prompt-based experiment using American blue-chip stocks.
- Quantitative assessment of sentiment accuracy compared to actual stock returns.
Excerpt from the Book
1.1 Introduction to the Topic
All big actors in the finance industry, namely banks, hedge funds, and market makers, have always tried to be ahead in technologies that could help them become more profitable. Especially in trading technology being one step ahead of competitors is crucial. For example, Spread Networks invested about $300 million to install a superfast fiber-optic line connection between New York and Chicago to save about three milliseconds in system response time and stay ahead of the competition. The finance industry is a rapidly evolving environment, and the convergence of technology and trading methodologies brings newfound opportunities.
The integration of large language models (LLMs) into finance opens doors to innovative applications. LLMs are capable of analyzing sentiment in text, which could be used to analyze news articles, earnings reports, or financial statements. Utilizing the advancements in artificial intelligence could potentially enable traders and investors to make more informed decisions. Especially using the newest technologies in combination with earnings releases opens opportunities for retail traders. Retail traders don't have the same financial background as big financial institutions, and therefore can't develop sophisticated trading algorithms. However, they might be able to compete by using an LLM that evaluates earnings releases to beat other market players. Of course, this topic is also very relevant to institutional investors for their own evaluation purposes.
Summary of Chapters
1 INTRODUCTION: This chapter contextualizes the importance of trading technology in modern finance and presents the core research question regarding LLM-based earnings trading.
2 EARNINGS TRADING STRATEGIES: Provides an overview of earnings seasons and existing pre- and post-earnings trading strategies used to capitalize on market volatility.
3 AMERICAN BLUE-CHIP STOCKS: Defines blue-chip stocks and explains their stability and historical data availability as the foundation for the experimental setup.
4 BASICS OF MACHINE LEARNING AND LARGE LANGUAGE MODELS: Covers the theoretical background of AI, Machine Learning, and the architectural concepts of LLMs.
5 EVALUATING THREE LLMS FOR SUITABILITY FOR EARNINGS TRADING: Compares BERT, RoBERTa, and GPT, ultimately selecting ChatGPT as the most suitable model for the experiment.
6 ASSESSING EARNINGS RELEASES WITH THE CHOSEN LARGE LANGUAGE MODEL: Details the methodology for stock selection, prompt engineering, and the execution of the sentiment analysis experiment.
7 ANALYSIS OF CHATGPT'S SENTIMENT RESPONSES: Analyzes the empirical results, comparing ChatGPT's predictions against actual stock movements for five specific companies.
8 FUTURE RESEARCH: Suggests potential improvements for further studies, including broader data integration and advanced AI fine-tuning.
9 CONCLUSION ABOUT THE LLM-BASED EARNINGS TRADING EXPERIMENT: Summarizes the thesis findings, acknowledging both the potential and limitations of using LLMs for stock market predictions.
Keywords
Earnings releases, Large Language Models, ChatGPT, Sentiment analysis, Financial trading, Blue-chip stocks, Machine learning, Prompt engineering, Investment strategies, Market volatility, Artificial intelligence, Stock market predictions.
Frequently Asked Questions
What is the core focus of this bachelor thesis?
The thesis investigates the capability of Large Language Models, specifically ChatGPT, to analyze corporate earnings releases and generate sentiment-based trading signals to potentially enhance investment returns.
Which specific financial instruments are analyzed?
The study focuses on five prominent American blue-chip stocks: Amazon, Coca-Cola, Walmart, Goldman Sachs, and The Walt Disney Company.
What is the primary goal of the research?
The primary goal is to determine if an LLM-based approach to sentiment analysis can accurately predict stock price movements following earnings announcements.
Which scientific methodology is applied in the study?
The research employs a quantitative experimental approach, testing various LLM architectures for suitability before executing a prompt-based sentiment analysis on historical earnings reports and comparing outcomes with actual market performance.
What topics are covered in the main section of the work?
The main sections cover the theory of earnings trading, the fundamentals of AI and neural networks, a comparative evaluation of LLMs, and a detailed performance assessment of sentiment accuracy across different timeframes.
Which keywords best characterize this research?
Key terms include Large Language Models, Financial Sentiment Analysis, Earnings Releases, Algorithmic Trading, and Blue-chip stocks.
How were the prompts for the AI developed?
The author utilized prompt engineering to simulate an impartial financial advisor, iteratively refining the instructions to ensure the model output provided a consistent 'Positive', 'Negative', or 'Neutral' sentiment classification.
What were the final conclusions regarding ChatGPT's performance?
The study found that while ChatGPT demonstrates some predictive ability, its performance varies across different stocks and timeframes, highlighting the need for critical evaluation and the inclusion of human oversight in real-world trading scenarios.
- Quote paper
- Timon Ortwein (Author), 2024, Assessment of Earnings Releases with a Large Language Model for Enhanced Returns with Five American Blue-Chip Stocks, Munich, GRIN Verlag, https://www.grin.com/document/1718356
