In this paper, we present FinSphere, a stock analysis agent, to overcome two major limitations: the absence of objective evaluation metrics for assessing the quality of stock analysis reports and the lack of depth in stock analysis that hinders the generation of expert-level insights. FinSphere consists of AnalyScore, a systematic evaluation framework for assessing stock analysis quality; Stocksis, a dataset curated by industry experts to enhance the stock analysis capabilities of LLM; and FinSphere itself, an AI agent that can generate high-quality stock analysis reports in response to user queries. Experimental results show that FinSphere achieves superior performance compared to existing agent-based systems as well as improved general and domain-specific LLMs with real-time data access and few-shot guidance. Combining real-time data feeds, quantitative tools, and fine-tuned LLMs with instructions, the integrated framework significantly improves the analysis quality and practicality for real-world stock analysis.
