Rolling Sharpe Alarm Thresholds
Setting rules that trigger a review or a de-risking action when a strategy's rolling Sharpe ratio drops below a chosen level, turning a vague sense that 'performance looks off' into a concrete, pre-agreed trigger.
Prerequisites: Sharpe Ratio
A single full-history Sharpe ratio tells you how a strategy has done overall, but it hides recent deterioration behind years of good performance. Rolling Sharpe recalculates the ratio over a moving window — the trailing three months, say — so that the number tracked day to day reflects recent behavior rather than an average that a strong first year can dominate for a long time afterward. An alarm threshold turns that rolling number into an actionable rule: if the rolling Sharpe falls below some pre-agreed level, a specific response — a review meeting, a size cut, a full stop — is triggered automatically rather than left to whoever happens to notice.
The hard part is choosing the threshold and the window sensibly, because Sharpe ratios estimated over short windows are noisy. A strategy with a genuine long-run Sharpe of 1.0 will regularly post three-month rolling Sharpes well below zero purely from sampling variation, even when nothing about the strategy has changed. Setting the alarm threshold too tight means constant false alarms that desensitize everyone to the warning; setting it too loose means real deterioration goes unnoticed for months. Good practice ties the threshold to the strategy's own estimated Sharpe ratio standard error, so the alarm is calibrated to how noisy that particular strategy's Sharpe naturally is, rather than using one blanket number for every strategy on the book.
A concrete example: a strategy with a backtested annual Sharpe of 1.5 is given a rule that a trailing 60-day rolling Sharpe below 0.0 triggers an automatic review, and below −0.5 triggers an automatic 50% size cut pending that review. Three months in, the rolling Sharpe dips to −0.2 — inside the review zone but not the cut zone. The review finds a temporary regime shift (a rates announcement disrupted the usual correlation structure the strategy relies on) rather than a structural break, and the strategy is left running at full size. Having the threshold pre-agreed meant the review happened calmly and on schedule, rather than after weeks of informal hand-wringing about whether things "felt off."
Rolling Sharpe alarm thresholds convert a vague sense that performance has slipped into a specific, pre-agreed trigger — but the threshold has to account for how noisy short-window Sharpe estimates naturally are, or the alarm will fire constantly on pure sampling variation.
Related concepts
Practice in interviews
Further reading
- Bailey and Lopez de Prado, 'The Sharpe Ratio Efficient Frontier'