COMPARING CHATGPT AND DEEPSEEK IN FINANCIAL FORECASTING: CROSS-REGIONAL EVIDENCE FROM TURKEY, EUROPE, AND THE UNITED STATES
DOI:
https://doi.org/10.7251/EMC2602322SKeywords:
Large Language Models (LLMs), Financial Forecasting, Earnings Prediction, Decision Support Systems, Confusion MatrixAbstract
This research compares the forecasting ability of two latest large language models ChatGPT (GPT-4) and DeepSeek-V3 on predicting corporate direction of earnings based on historical financial reports alone. The panel data of 15 Turkish, European, and American companies operating in the banking, technology, and industrial sectors for 2022-2024 have been considered. Projections were made with a consistent prompt style and evaluated for direction accuracy, confusion matrices, interpretive insight scores, and sector-region performance.
Findings suggest that while ChatGPT offers greater qualitative explanation and performs better in unstable, recovery-based scenarios, DeepSeek offers greater accuracy, stability, and performance in stable, regulation-based environments. The paper brings into focus critical trade-offs between interpretability and reliability and suggests that the use of hybrid models can offer the best solution to financial forecasting.
These findings oppose Efficient Market Hypothesis assumptions and highlight the need for decision-support systems that integrate AI interpretability, domain expertise, and responsible model selection. The research offers prescriptive advice to analysts, institutions, and developers who aim to incorporate LLMs into financial decisions.
