The GARCH model is one of the most powerful trading frameworks used daily by real quant analysts. My recent article explains exactly how I used Claude Code to automate this GARCH quant trading strategy. Here's the full architecture behind it, with real numbers, and how I got Claude to build it for me: What GARCH actually measures It's a Nobel Prize-winning calculation, originally developed as "ARCH" by Robert F. Engle in 2003, then extended by his student Tim Bollerslev into the GARCH model quant desks still run today. The whole point: market volatility isn't random. It's predictable. GARCH is how you measure and forecast it. The calculation (only 3 inputs) GARCH forecasts tomorrow's volatility as the sum of: → Base level movement: every asset carries a baseline amount of volatility that never really changes. Bitcoin's baseline is higher than the S&P's, always has been. → Yesterday's shock: how hard did the market actually move in the last session? → Yesterday's volatility level: where was volatility already sitting before that move happened? Add those three together, and you get tomorrow's violence score. Real example from backtested data On a typical day, there's roughly a 10% chance of a violent move in BTC. The day after a shock like the FTX collapse, that probability triples to roughly 30%. As of writing, the Nasdaq's violence score is running at 30% annualised. Turning that violence score into an actual position size Drawdown limit ÷ violence score = your position size multiplier Institutional desks typically cap annual drawdown at 15%. So: 15/30 = 0.5x. On a $10,000 baseline position, that would mean your postion size should be $5,000 to manage risk. Keep in mind, with this framework, you're not predicting direction; you're controlling exposure to risk that's already been measured. The important caveat GARCH tells you how much the market might move. It doesn't tell you which direction. It's a volatility forecast for managing portfolio risk, not a directional trade signal. If a genuine crash erupts out of a calm period, this framework won't catch it; you'd take it at full size. Does it actually work? I backtested the same EMA cross strategy two ways across 15 years of BTC data. Same entries, same exits, same signals. → Fixed position sizing: $17,957 final equity → GARCH position sizing: $21,205 final equity The GARCH version also took fewer drawdowns and less overall market risk to get a better result. Full breakdown of the entire quant strategy, plus the open-source Claude Skill and GitHub repo, in the article. Article pinned on my profile.(Miles Deutscher)