Large Language Models in Accounting: Automating Financial Statement Analysis

We evaluate the ability of large language models (LLMs) to analyze financial statements and generate investment insights comparable to professional financial analysts. Using GPT-4 and domain-specific fine-tuned models, we find that LLM-generated analyses achieve 72% accuracy in predicting earnings surprises, approaching the 78% accuracy of top-quartile human analysts. The models excel at processing standardized financial data but struggle with industry-specific contextual factors.