Drawdown, variance and sample size
Why a profitable process can lose for long stretches—and why short winning streaks are weak evidence.
Variance is not a footnote
Binary outcomes are noisy. Even when a probability estimate is directionally useful, wins and losses can cluster in ways that feel meaningful but are compatible with chance.
This is why screenshots of a winning week, a perfect accumulator or a recent streak are not a reliable audit of a forecasting system.
Drawdown measures the difficult path
Maximum drawdown tracks the largest peak-to-trough fall in a sequential unit ledger. It exposes the path hidden by a final ROI number.
The result is sample-specific and depends on staking assumptions. It should be calculated from the complete append-only series, including losses and voids.
Maturity requires more than a threshold
A larger sample reduces some uncertainty but does not repair data leakage, changing methods or selective publication. Cohort size and process integrity must mature together.
- Use fixed unit accounting.
- Show the full sequence.
- Version methodology changes.
- Avoid annualising a tiny sample.
Sources
Direct links are preserved so the editorial reasoning can be checked independently.