Drawdown

The decline in account equity from a peak to a subsequent trough. Reported in dollars (DD$) and as a percentage of the peak (DD%). The single most important risk metric for any leveraged strategy.

In plain English

Your account equity rises and falls over time. A drawdown is the worst stretch — the largest "peak-to-valley" drop measured along the equity curve. If your account peaked at $1,000 and then dropped to $700 before recovering, that's a $300 drawdown, or 30%.

Drawdown matters more than final PnL for one simple reason: you have to live through the drawdown to collect the PnL. A strategy that ends up making $1,000 after first losing 95% of your capital is mathematically a profit but practically a catastrophe — you would have stopped trading it, run out of margin, or been liquidated long before the recovery.

Formula / mechanic

For an equity curve E_t over time, the running peak is P_t = max(E_0, ..., E_t). Drawdown at time t is:

DD_t  = P_t − E_t        (in dollars)
DD%_t = (P_t − E_t) / P_t  (as a percentage)

The "max drawdown" reported in analytics is max(DD_t) over the full backtest window.

Why it matters for this fleet

Many high-leverage strategies in this fleet have drawdowns deep enough that a real account would have been wiped out (or at minimum, margin-called and force-deleveraged) before any recovery arrived. The simulator continues taking trades past the point a real broker would have stopped — so headline returns on high-drawdown strategies are fictional.

The honest deployable criterion: max drawdown shallow enough that your real capital survives it. Aspirationally: max DD% < 20%.

Examples from the live fleet

The cleanest way to see drawdown's stakes is to hold the signal fixed and vary only the leverage (the multiplier on position size — 2× means twice the exposure, twice the swing).

• id523 — EMA 21/50 · SOL · 1h · 2× · long: max drawdown −9.9%. Survivable. The worst peak-to-trough dip is a tenth of the account — uncomfortable but recoverable.
• id522 — EMA 21/50 · ETH · 4h · 50× · long: max drawdown −60.9%. The same family of idea, but at 50× the account lost more than half its value at its worst point. A real trader would very likely have capitulated or been force-deleveraged long before any recovery.

The leverage cliff

Across the whole fleet, the median max drawdown climbs steeply with leverage — drawdown is the price you pay for the amplified returns leverage promises:

By 10× the typical strategy has already lost most of the account at its worst point; by 50× it is all but wiped. This is why drawdown — not headline return — is the metric that decides whether a leveraged strategy is deployable.

Why the dollar amount matters more than the percentage

The same DD% means very different things on different account sizes. A 50% drawdown on a $1M account is recoverable (managed account, lots of cushion). A 50% drawdown on a $200 account is psychological devastation and probably ends in capitulation. **Always cross-check DD$ against your actual deployable capital.**

Why drawdown is path-dependent

The same set of trades arranged in a different order produces different drawdowns. A strategy that takes 7 losses then 3 wins has a different DD than one that takes 3 wins then 7 losses. The reported max DD is the realized drawdown from this particular historical ordering — not the worst case. See path dependence.

How to defend against drawdown

• Reduce leverage. Halving leverage halves max DD$ on the same trades.
• Add stop-losses. Caps the loss on individual trades; lowers max DD at the cost of cutting some recoveries early.
• Position-size on remaining equity. Reduces position size after each loss so that further losses shrink in absolute terms.
• Run multiple strategies in parallel. If their drawdowns are uncorrelated, portfolio max DD is much smaller than the sum of individual max DDs.
• Pick lower-DD strategies in the first place. See risk of ruin.

Buy-and-hold drawdown as a benchmark

Buy-and-hold drawdown (peak-to-trough loss of a passive 1× holder over the same window) is a meaningful, direction-neutral benchmark for a strategy's own drawdown. The question "did your strategy suffer less than just holding?" can be asked of any strategy — long, short, or bidirectional — because capital protection is direction-independent.

Contrast with buy-and-hold PnL (see alpha), which is direction-dependent: a short strategy in a bull market looks bad vs buy-and-hold return simply because the benchmark assumed a long position.

The catch: a short strategy in a sustained bull run can have worse drawdown than buy-and-hold (the persistent rally against the short is the drawdown). So buy-and-hold drawdown as a benchmark doesn't always favor the active strategy — but the question is still universally meaningful.

The engine already has the buy-and-hold equity curve (computeBenchmark) but does not yet extract its max drawdown. Captured as SEED-025. See risk adjusted return for the full gap description.

Related

• drawdown duration — how long equity stays below the peak; the psychological endurance dimension depth alone misses
• liquidation
• path dependence
• risk of ruin
• leverage
• sharpe ratio
• buy and hold
• risk adjusted return

Sources

• wiki/qa-sessions/2026-05-17-session.md#q9 (first formal entry; referenced throughout earlier sessions)
• wiki/qa-sessions/2026-06-28-session.md#q9 (buy-and-hold drawdown as direction-neutral benchmark)
• /api/analytics maxDrawdown (%) and maxDrawdownDollar fields