Understanding drawdowns: what every algo investor should know.

A drawdown is the gap between the high-watermark of an equity curve and any subsequent low. It is the most honest number in any performance report — and the one investors most often want to forget about. There is no trading system that produces returns without producing drawdowns. The questions to ask are: how deep, how long, and how often.

How deep.

Peak drawdown — the largest single peak-to-trough gap in the record — sets the expectation for the worst case the system has historically tolerated. If a strategy advertises an 8% peak drawdown, the realistic expectation is that future drawdowns may equal or exceed that figure under sufficiently bad conditions. Anyone telling you 'this will never happen again' is selling, not reporting.

How long.

Time-to-recovery is more important than depth. A 6% drawdown that recovers in two weeks is forgettable. A 6% drawdown that takes nine months to recover is a different category of pain — it tests an investor's relationship with the manager more than it tests the manager's models.

The algorithm doesn't feel the drawdown. The investor does. That asymmetry is most of risk management's actual job.

How often.

A clean equity curve isn't a curve with no drawdowns. It is a curve where drawdowns occur on a predictable cadence and stay within an envelope. Our monthly returns heatmap is designed for this read — at a glance you can see whether the system has produced losses on a frequency consistent with its design, or whether something has changed.

When a drawdown is telling you something is wrong.

There are two kinds of drawdowns. The first is statistical — the model is correct, the regime is unfavorable, and the curve recovers as the regime mean-reverts. The second is structural — something in the model is no longer right. The difference is visible in the trade ledger long before it is visible in the equity curve. A rising loss-per-trade, a falling hit rate, or a clustering of losses in a single asset all warn that the cause is structural, not statistical. A good firm reacts to the second kind. A bad firm hopes through it.

Our risk envelope is built around the assumption that the second kind happens. The kill-switch tier doesn't ask permission; it acts.