daily_stock_analysis review: how reliable is LLM-based stock analysis?
Open-source daily_stock_analysis brings LLM-powered analysis to A-shares, Hong Kong and US markets. Setup, cost, benchmarks and pitfalls covered in this hands-on review.
daily_stock_analysis is an open-source project with over 3,900 GitHub stars that pulls market data from A-shares, Hong Kong and US exchanges, runs LLM-based technical and news sentiment analysis, and pushes results via email or DingTalk automatically. The review walks through installation, LLM provider configuration in config.yaml (supporting OpenAI, DeepSeek, and Tongyi Qianwen), and how model choice drastically affects cost—switching from GPT-4 to gpt-4o-mini cut the price of analyzing 20 stocks from roughly $4 to about $0.3.
Benchmarks on a 2-core/4GB server measured data-fetch time, LLM analysis time, and token usage across batches of 10, 20, and 50 stocks, showing smooth performance up to 20 stocks but significant slowdown beyond that. The author also documents real-world issues and fixes: an akshare version conflict, missing data for obscure Hong Kong tickers, environment variable problems breaking crontab scheduling, and the LLM occasionally hallucinating news facts—addressed with a simple validation function. A free GitHub Actions workflow is provided for zero-cost daily scheduling.
Compared to StockSharp and JoinQuant, the tool's standout feature is native LLM integration, something neither competitor offers, though it lacks a backtesting framework and isn't suited for quant strategy development. The overall rating lands at 4.2/5, making it a good fit for retail investors with basic Python skills, developers exploring LLM-finance integration, and content creators generating daily market commentary—but not ideal for beginners or serious quant traders.