Run monthly A-share stock analysis using ensemble models to identify which stocks to avoid holding.
Add AI analysis buttons directly to Eastmoney stock pages using the Chrome extension for quick signal checks.
Retrain the LightGBM or Kronos model on your own A-share dataset using the provided Python scripts.
Get plain-language summaries of what the quantitative signals say about a specific stock via the language model component.
Requires setting up the Chrome extension, Node.js native host bridge, and Python ML environment with model weights before analysis works end to end.
This is a personal research project for analyzing Chinese A-share stocks on a monthly basis. The stated goal is avoiding losses rather than chasing profits. It combines several independent analysis methods, including statistical factors, machine learning models, and an AI language model, and tries to blend them intelligently based on current market conditions. The core idea is that no single model works well in all market environments. The system first classifies what kind of market it is currently in, such as a trending market, a volatile one, or a sideways one, and then routes to the signals that have been validated specifically for that condition. The regime detection uses three independent methods that vote, so no single failure brings down the whole system. The AI language model is kept completely separate from the decision logic: it only generates plain-language explanations of what the other signals are saying, it does not influence the actual scoring. The project went through a methodological research cycle that uncovered some significant problems in earlier versions. Excluding delisted stocks from the evaluation pool inflated apparent returns by 8.4 percentage points, so the team rebuilt the dataset to include them. The original scoring system also had a bug that made predicting neutral outcomes every single time the optimal strategy. After fixing both issues, the validated signals were: a transformer model called Kronos (reliable across most conditions), momentum and a LightGBM model (work well in bear markets, flip negative in bull markets), and the language model (predicts neutral 70% of the time, but shows a notable edge when it does commit to a strong call). The project can be used as a Chrome extension that adds analysis buttons directly to Eastmoney stock pages, as a Node.js command-line tool for batch processing, or through Python scripts for model training and research. A native host bridges the extension to local compute for database access and ML inference. This is described explicitly as a personal research project with no guarantees of continued updates or backward compatibility. It does not execute trades or connect to brokerages. The license is MIT.
← yuu-ramsey on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.