Strategy drift

What drift looks like in the data

Drift is visible before it is profitable or unprofitable. An account that held positions for days starts closing within hours; one that traded three majors starts trading long-tail perps; sizing that was stable for months begins to vary sharply.

None of these are inherently bad decisions by the trader. They are simply changes that make the historical record a weaker description of current behaviour.

• Holding-period shift, in either direction.
• Market migration, especially into thinner books.
• Sizing instability, or a step change in typical notional.
• New behaviour such as adding to losers where none existed before.

Why PnL is a lagging detector

A drifting trader can post good results for a while — often that is why they changed. Waiting for the loss to arrive means the follower has already funded the experiment.

Behavioural detection is deliberately more sensitive than performance detection, which means it produces false alarms. The sticky basket and the strike system exist precisely so a suspected drift has to show up as a repeated soft issue rather than triggering an abrupt exit on one bad session.

How the system responds

Scores are recomputed continuously, so drift that degrades the measured dimensions lowers the score and therefore the allocation, but a lower score alone does not remove a leader from the sticky basket. Drift that produces repeated soft-issue strikes (three strike-days) or an emergency condition triggers replacement, with their sub-account unwound independently of the rest of the basket.

Past performance is not indicative of future results. Perpetual futures are leveraged instruments and carry a substantial risk of loss, including the loss of your entire position.