Professional trader Miles Deutscher has revolutionized risk management by automating the famous Nobel Prize-winning GARCH model using the AI agent Claude Code. By his own admission, the entire system was assembled by artificial intelligence: the trader only set the task, and the neural network itself turned complex mathematics into a working tool.
What the model measures and how AI built it
GARCH (Generalized Autoregressive Conditional Heteroskedasticity) is one of the most powerful quantitative models used daily by professional quants. Its founder, Robert Engle, received the Nobel Prize in 2003 for its development, and his student Tim Bollerslev expanded the model to GARCH, which is still used by quantitative analysis departments worldwide.
The key idea of the model is that market volatility is not random but can be predicted. Deutscher explains that GARCH allows measuring and forecasting fluctuations in advance, based on just three indicators:
- The base level of asset fluctuations (for Bitcoin, it is traditionally higher than for the S&P 500 index).
- The strength of yesterday's market movement.
- The level of volatility at which the market was before this movement.
The sum of these three values provides a forecast of fluctuations for tomorrow. It is important to understand: the model shows how much the market might move, but does not indicate the direction. It is a portfolio risk management tool, not a buy or sell signal.
The role of AI in this project proved crucial. Deutscher emphasizes that he did not program the system himself — he described the task in natural language, and Claude Code handled all the technical implementation.
How the forecast translates into position size
The trader uses the obtained fluctuation estimate to calculate position size. Institutional departments typically limit annual drawdown to 15%, and all subsequent mathematics is built from this limit. If the fluctuation estimate for an asset is, say, 30% annually, the drawdown limit is divided by this figure — and the position multiplier becomes 0.5x. With a base position of $10,000, this means entering at $5,000.
Deutscher tested the model's real benefits on historical data. He ran the same strategy on a moving average crossover in two ways on a fifteen-year dataset for Bitcoin — with identical entries, exits, and signals. With a fixed position size, the final capital in the test was $17,957. With position sizing based on the GARCH model, the result was higher — $21,205.
According to the trader, the version with the Nobel model not only generated more profit but also went through the process with smaller drawdowns and overall risk. However, he warned of an important limitation: during a sharp crash after a calm period, the model will not work, and the position will be opened in full.
Expert opinion: The integration of classical financial models with modern AI agents opens new horizons for retail traders. Previously, GARCH was the prerogative of institutional funds with development teams; now, any savvy trader can gain access to advanced risk management. However, one should not forget: no model, even a Nobel Prize-winning one, guarantees protection against extreme market events.