At the intersection of fundamental science and practical trading, truly interesting tools emerge. Trader Miles Deutscher has introduced an automated trading strategy built on the GARCH (Generalized Autoregressive Conditional Heteroskedasticity) model — a mathematical framework that earned a Nobel Prize. The most remarkable part: the entire system, from concept to implementation, was assembled by the artificial intelligence Claude Code, not a human.

What is GARCH and how did AI build this model

GARCH is one of the most powerful quantitative models that professional quant analysts use daily. Developed by Robert Engle (Nobel Prize 2003) and expanded by his student Tim Bollerslev, it underpins the risk management of many institutional departments.

The key idea of the model is that market volatility is not random — it is predictable. The calculation relies on just three indicators:

  • Base level of fluctuations — a constant for each asset. For Bitcoin, it is traditionally higher than for the S&P 500 index.
  • Strength of yesterday's market movement — how much the price changed in the previous period.
  • Volatility level before that movement — historical context.

The sum of these three values provides a forecast of fluctuations for tomorrow. Deutscher emphasizes: he did not program the system himself. He only described the task, and Claude Code turned the math into a ready-to-use working tool.

How volatility forecast translates into position size

Deutscher uses the obtained fluctuation estimate to calculate position size. In the institutional environment, it is common to limit annual drawdown to 15%. All the math is built from this limit.

Example: if the fluctuation estimate for an asset is 30% annually, the drawdown limit is divided by this figure — the position multiplier becomes 0.5x. With a base position of $10,000, this means entering at $5,000.

This approach does not predict price movement — it controls exposure to already measured risk. As Deutscher noted, at the time of writing, the fluctuation estimate for the Nasdaq index stood at 30% annually.

Backtesting on historical data: impressive results

To test effectiveness, the trader ran the same strategy on a moving average crossover in two ways over a fifteen-year dataset for Bitcoin — with identical entries, exits, and signals.

Results:

  • With a fixed position size, the final capital was $17,957.
  • With position calculation using the GARCH model, the result was higher — $21,205.

The version with the Nobel model not only generated more profit but also navigated the path with smaller drawdowns and overall risk. However, Deutscher warns: during a sharp crash following a calm period, the model will not work — the position will be opened in full.

My expert opinion: Using GARCH combined with AI automation is an elegant hybrid of academic rigor and modern technology. But do not forget: the model measures volatility, not direction. It is a powerful risk management tool, not a crystal ball. The strategy's success depends on discipline and understanding its limitations.