AI in the Order Flow: What Machine Trading Is Doing to Volatility

Machine learning has been in markets for years, but the current generation of adaptive execution and model-driven strategies behaves differently from its rule-based ancestors, and researchers are working out what that means for volatility itself.

Early findings are mixed in instructive ways. Adaptive strategies appear to dampen volatility in normal conditions, adjusting quickly to absorb flow imbalances, while contributing to sharper moves when many models respond to the same signal simultaneously. Calm most days, and a faster crowd on the days that matter.

Regulators are asking measurement questions before policy ones: how much flow is model-driven, how correlated are the models, and whether existing circuit breakers match the speed of the herding they are meant to interrupt. Exchanges, holding the data, have become reluctant referees in the debate.

Practitioners mostly shrug at the framing. Every generation of market technology, they note, was accused of breaking volatility until it became the baseline against which the next generation was accused. The honest answer is that the market is running the experiment on itself, and the results publish in real time.