Who it is forA quant learner who wants to test a factor idea without fooling themselves with overfitting.
The problemBacktests that reuse the same data for tuning and testing look great and fail live.
The outcomeIC, ICIR, turnover and decay metrics with a one-time hidden OOS test, logged with fingerprints in an offline HTML report.
What you need first
- Python 3.11 or later
- A local copy of the repo (inferred: git clone https://github.com/ceyyy427/finathink.git)
- No broker, live trading account or real vendor credentials are needed for the offline sample run
Steps
- Create a venv and run
python3 -m pip install -e '.[dev]' - Start the local app with
python3 scripts/run_local_app.py --sample --port 8765 - Describe your hypothesis in the factor workbench so it maps to a fixed DSL template
- Run training, validation and the one-time hidden OOS test
- Read IC, ICIR, turnover and decay, and keep the proposal only if OOS holds up
Prompt for your agent
Using ceyyy427/finathink with the sample data, turn the hypothesis <idea> into a factor proposal, run the backtest with the hidden OOS test, and summarize IC, ICIR, turnover and decay. Paper trade only.
Caveat. Not financial advice: paper trade first, and remember you can lose money. The README is in Chinese.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for ceyyy427/finathink and "Backtest a factor idea with a hidden out-of-sample test"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer