| Abstract [eng] |
The pursuit of financial independence has become increasingly relevant in the digital age, as internet accessibility provides vast amounts of information and new investment opportunities. Over the last decade, significant advancements in Artificial Intelligence (AI) have transformed daily tasks. However, its application in achieving consistent investment success remains a complex challenge. While existing literature explores various AI implementations in the financial sector, a definitive and universally successful strategy for AI-driven investing has yet to be established. The primary objective of this thesis is to develop and evaluate an automated stock trading system based on Large Language Model (LLM) agents. The research begins with a comprehensive analysis of investment theories, AI literature, and the history of trading automation. In the practical part of the work, six distinct LLM-based prototypes were developed and tested through real-time paper trading. The performance of these prototypes was then evaluated using key financial metrics, including cumulative return, Sharpe ratio, and maximum drawdown. The literature review revealed a significant gap in the evaluation of LLM efficiency within active trading environments. The experimental results of this study demonstrated that LLM-based agents can effectively navigate financial markets, with "DeepSeek" and "Gemini" prototypes achieving the most efficient returns and demonstrating high risk-adjusted performance. These findings suggest that LLM-based automation can serve as a viable tool for modern investment strategies. This paper consists of 62 pages, 8 tables, and 23 illustrations. |
