Capital jumps and drawdown

The measurement everyone quotes

Max drawdown is the largest peak-to-trough decline in a series. Applied to an equity curve it answers a question that matters more than return: how bad did it get on the way here? Two traders with the same 30-day return are not equivalent if one drew down 8% and the other 55%, because the second one is far closer to the version of the outcome where they do not survive.

That is why drawdown, not return, is the natural gate on a leader. A return threshold asks whether someone made money. A drawdown threshold asks whether their process is survivable. Only the second one generalises.

The problem is what you feed into the calculation.

Why an on-chain account-value curve is not an equity curve

A trader's account value on Hyperliquid changes for two completely different reasons: trading (fills, funding, liquidations) and capital movements (deposits and withdrawals). A naive drawdown calculation cannot tell them apart, because both appear as the same thing — the line went down.

Consider a trader who grows an account from $100,000 to $180,000 over three weeks and then withdraws $70,000 to a cold wallet. The account-value curve falls from $180,000 to $110,000. A naive calculation reports a 39% max drawdown and, on a 35% threshold, flags the single best-performing wallet in the set as a risk failure. Nothing bad happened. They took profit off the table, which is prudent behaviour.

The inverse is just as wrong. A trader loses 30% over two weeks, then deposits fresh capital equal to the loss. Account value returns to its old high, so the naive curve shows a shallow dip and a full recovery. The trading was bad and the risk profile is unchanged; the number has been rescued by a wire transfer.

Professional traders move capital constantly — profit-taking, margin management, moving size between venues, rotating to cold storage. So this is not an edge case affecting a few wallets. It is a systematic distortion that biases the measurement against exactly the disciplined, profit-taking behaviour you want to select for.

• A withdrawal reads as a loss the trader never took.
• A deposit reads as a recovery the trader never earned.
• The bias is not random: it penalises disciplined profit-taking and flatters top-ups.

What jump-adjusted drawdown does

The fix is to detect the capital movements and neutralise them before measuring. Conceptually: walk the series, identify discontinuities that are not attributable to trading, and adjust the curve so the sequence measures the same book of trades without the transfer. What remains is a return series that answers 'how did the trading go' rather than 'how did the balance move'.

Mechanically this is closer to how a fund reports performance than how a bank statement reads. A fund with inflows and outflows does not report a drawdown when an investor redeems; it measures the return of the strategy per unit of capital. Jump adjustment applies the same discipline to a wallet, using the capital movements that are visible on-chain.

Applied to the earlier examples, the profit-taking trader shows a modest drawdown consistent with their actual trading path, and the topping-up trader shows the full 30% decline they genuinely produced. The ordering between them inverts relative to the naive calculation — which is the point.

How this feeds the replacement policy

Drawdown is one of the soft issues in the live replacement policy, and the threshold is a 30-day trading max drawdown above 35% measured on jump-adjusted equity. Two details in that sentence carry weight.

First, 'jump-adjusted' means a leader is never removed because they withdrew profits, and never protected because they deposited into a losing streak. Without the adjustment, the policy would fire on capital-management behaviour and miss real risk, which would make the whole gate worse than not having one.

Second, 'soft issue' means tripping the threshold does not remove a leader on its own. It records at most one strike for that UTC calendar day. Three separate strike-days trigger a replacement, and a clean evaluation day resets the counter to zero. A single volatile fortnight inside a strategy that still works does not end a leader's membership.

The same reasoning drives the return test. The 30-day return gate is measured from realised PnL rather than raw account-value change, at a -15% threshold, for exactly the reason described above: an account-value delta over a window containing a withdrawal is not a return.

• Drawdown gate: 30-day trading max drawdown above 35%, on jump-adjusted equity.
• Return gate: 30-day PnL-based ROI below -15%, not account-value change.
• Both are soft issues: max one strike per UTC day, three strike-days to replace, clean day resets.

How to read this on a trader page

The same problem affects the ROI figures shown on public rankings, which is why trader pages on this site surface how a return was estimated and whether capital jumps were detected in the window, rather than presenting one confident number.

When a wallet shows detected capital jumps, treat any naive account-value-derived figure for that window as unreliable, and prefer the PnL-based view. When a ranking elsewhere shows a spectacular ROI with no indication of how it was computed, assume it is account-value-derived and treat it as unverified.

This is the practical version of the general rule: on-chain data is transparent, but transparency is not the same as being correctly interpreted. The raw series is public; the adjustment is where the judgement lives, and it is worth asking any provider how they do it.