Trader Miles Deutscher has made a breakthrough in automating trading strategies by tasking artificial intelligence with building a full-fledged system based on the GARCH model. All the work of integrating the mathematical framework into a working tool was carried out by Claude Code — the trader himself only described the task.
The GARCH model is one of the most powerful quantitative tools, used daily by professional quant analysts. Its predecessor, ARCH, was developed by Robert Engle in 2003, for which he received the Nobel Prize. Later, his student Tim Bollerslev expanded the model into GARCH, which remains the standard for measuring and forecasting volatility.
How volatility forecasting works
The main idea of GARCH is that market volatility is not random — it can be predicted. The calculation relies on just three indicators:
- Base level of fluctuations — a constant inherent to each asset. For Bitcoin, it is traditionally higher than for the S&P 500 index.
- Strength of yesterday's market movement — the reaction to the last significant impulse.
- Previous volatility level — the state of the market before that movement.
The sum of these three values provides a forecast of fluctuations for the next day.
It is important to understand: the model does not predict the direction of the price. It is a portfolio risk management tool, not a buy or sell signal. Deutscher emphasized that the volatility forecast allows calculating the position size, not indicating where the market will go.
Practical example: from forecast to position size
Institutional departments typically limit the annual portfolio drawdown to 15%. If the estimated fluctuations for an asset are 30% annually, the drawdown limit is divided by this value — and the position multiplier becomes 0.5x. With a base position of $10,000, this means entering at $5,000.
Deutscher tested the strategy on a fifteen-year dataset for Bitcoin. With a fixed position size, the final capital amounted to $17,957. With the position calculated using the GARCH model, the result was higher — $21,205. Moreover, this path was traversed with smaller drawdowns and lower overall risk.
However, there is a limitation: during a sharp crash following a calm period, the model may fail, and the position could be opened in full.
Expert opinion: Integrating the Nobel-winning GARCH model via AI is a logical step for retail traders seeking institutional-level risk management. However, relying solely on volatility without considering the direction of movement is a strategy for capital preservation, not aggressive growth. In the long term, this approach is justified, but during periods of structural market shifts, it may not prevent losses.