All projects

ETF rotation bot

Does a simple weekly ETF rotation hold up on a paper account?

Why I built it

No machine learning makes the decisions. The rules are simple on purpose so I can explain every trade; an LLM only writes a short summary after each rebalance.

What it does

A weekly rules-based ETF rotation on an Alpaca paper account, packaged with Docker and monitored from a Streamlit dashboard.

Under the hood

  1. Market regime

    Every Friday at 15:45 New York time: bearish if SPY is below its 200- and 50-day averages and more than 5% under its 60-day high.

  2. Defensive or offensive

    Bearish: gold, commodities and Treasuries. Bullish: the 7 best of 24 assets by trend quality and return over volatility.

  3. Volatility scaling

    Offensive exposure goes from 97% in calm markets to 30% in turbulent ones, and is cut again if the account falls 10% below its peak.

  4. Hard limits

    12% per asset, 28% per sector, 2% cash always kept, no leverage; orders in dollar amounts, skipping any position within 4% of target.

  5. An honest backtest

    Weekly since 2016 with fees and a stress case, and the README lists its flaws: look-ahead in the asset list, approximated Bitcoin ETFs, a simplified copy of the live rules.