When momentum signals persist far longer than models predict

In the winter of 2021, a momentum signal emerged in a sector that most analysts had already dismissed as structurally impaired. The conventional models assigned it a six-week lifespan. Eighteen months later, it was still accelerating. This was not an isolated anomaly. Across multiple asset classes and time periods, certain signals have demonstrated a stubborn persistence that challenges the neat decay curves drawn by quantitative researchers. The question is not whether these exceptions exist, but what they reveal about the limitations of standard measurement frameworks.

One explanation lies in the feedback loops created by institutional behaviour. When large mandates shift direction slowly, their accumulated flows can sustain a price trend long after the original catalyst has faded. The signal persists not because the underlying thesis remains valid, but because the mechanics of position adjustment create their own momentum. Analysts who measure only price and volume miss the structural plumbing beneath the surface, where pension rebalancing, risk-parity flows, and volatility-targeting strategies extend trends beyond their natural lifespan.

Another factor is the changing nature of market participation itself. The rise of systematic strategies has created a layer of automated response that did not exist when many canonical momentum studies were conducted. These programmes do not fatigue, do not reconsider, and do not take profits for emotional reasons. They can sustain directional pressure in ways that human traders never would, producing persistence patterns that historical data cannot adequately explain. The models assume a certain ecology of participants, and that ecology has shifted beneath their feet.

The uncomfortable synthesis is that signal decay is not purely a property of the signal itself, but of the environment in which it operates. Analysts who treat half-life as an intrinsic characteristic are making a category error. The same pattern that expires in six weeks during one regime might persist for eighteen months in another, not because the mathematics changed, but because the participants and their constraints did. This recognition does not make prediction easier, but it does make humility more necessary.

For the practical investor, the lesson is to build position management around scenario ranges rather than point estimates. Instead of asking when a signal will die, one might ask what conditions would cause it to persist, and whether one's portfolio could survive that extension. The models provide a starting point, not a verdict. The final arbiter remains the market itself, which has never felt obliged to conform to anyone's expectations, however carefully constructed.