Hyperliquid Risk Management: The Complete Guide (Position Sizing, Margin, Liquidation)

Hyperliquid removed most of the classic excuses. Execution is fast, the book is deep, custody is yours, and fills land on-chain where you can verify them. What it did not remove — what no venue can remove — is the part of trading that actually determines whether an account compounds or dies: the risk decisions.

Accounts on perp venues rarely die from bad market calls. They die from a good call sized 10x too large, from a stop that was a mental note instead of an order, from funding quietly bleeding a crowded position, from two 'different' trades that were the same trade, and from one revenge entry after a losing streak. Every one of those is a risk-management failure, and every one is preventable with arithmetic that fits on an index card.

This is the complete system: the five numbers that define every trade, how Hyperliquid's margin machinery actually works, where liquidation really sits, what funding costs, how exits should be placed, and the portfolio-level rules on top. Each section links to a deeper guide where one exists. None of it is exotic — the edge is not in knowing it, but in applying it on every single trade.

Strip any perp trade to its skeleton and five numbers remain. Decide them in this order and most risk mistakes become impossible:

• Account risk — the fraction of your equity this trade may cost if the stop hits. Serious answers live between 0.5% and 2%. This is the only number you choose freely; everything else derives from it.

• Stop distance — how far away invalidation is: the level where the trade idea is objectively wrong. This comes from market structure, not from what feels tolerable.

• Position size — the output of the first two: size = (equity × account risk) ÷ stop distance. Not an input. Never an input.

• Leverage — bookkeeping that falls out of size and the margin you allocate. Leverage is how the position is financed, not how big it should be.

• Liquidation distance — where forced closure sits, which must be several times further than the stop. If it isn't, the position is over-financed regardless of how good the idea is.

Most blown accounts run this list backwards: they pick a leverage that feels exciting, derive a size from it, and discover their stop and liquidation distances afterwards — usually live.

The formula deserves its own line, because it is the entire discipline in one expression:

• Position size (notional) = (equity × risk fraction) ÷ stop distance

A $10,000 account risking 1% on a BTC setup whose invalidation sits 4% away: ($10,000 × 0.01) ÷ 0.04 = $2,500 of notional. If the stop is tight — 1% — the same 1% risk supports $10,000 of notional. Same dollars at risk, four-times-different position, and the difference came entirely from the geometry of the stop.

Notice what never appeared in the calculation: leverage. On Hyperliquid you might finance that $2,500 position with $500 of isolated margin (5x) — fine, because the liquidation sits ~19% away while your stop sits at 4%. The 5x changed the financing, not the risk. The risk was fixed the moment you chose 1% and the stop level.

This compresses further into the habit that separates surviving traders from statistics: decide where you are wrong before deciding how big you are. The full treatment — worked examples at Hyperliquid's tiers, the classic sizing mistakes, volatility adjustment — is in the position sizing guide.

Hyperliquid gives every position one of two relationships with your collateral.

Cross margin (the default) pools your entire perps balance behind all cross positions. Maximum cushion per position, but shared fate: a deteriorating position drains the same equity that supports every other one, and a true blow-up takes the book down together.

Isolated margin fences a fixed amount behind one position. If it fails, the loss is capped at that fence — nothing else is touched. You can add margin to an isolated position later, or remove it (bounded so the position stays properly collateralized).

The practical split follows from what each mode is good at. Core positions you manage attentively, with sane leverage and correlated exposure you understand, sit reasonably in cross — the shared cushion is efficiency. Anything convex, experimental, or high-leverage belongs in isolation: tail-risk trades, breakout attempts, anything where being catastrophically wrong is on the menu. Isolation converts 'wrong' from contagion into a fixed, pre-paid cost.

One asymmetry worth internalizing: cross margin fails loudly at the account level, isolated fails quietly at the position level. Choose which failure you would rather explain to yourself later.

Hyperliquid's liquidation machinery has three facts that every position should be designed around.

First, the floor: maintenance margin is half of the initial margin at the asset's maximum leverage — 1.25% of notional for BTC (40x max), 2% for ETH (25x max). Your liquidation distance is roughly your margin fraction minus that floor: at 10x, about 8.8% adverse; at 20x, about 3.8%; at 40x, about 1.25%. Against BTC's routine 3-5% daily range, high leverage is a bet that nothing normal happens.

Second, the reference: liquidations trigger on mark price — a median that blends the CEX oracle, Hyperliquid's own book, and external perp mids. A wick on Hyperliquid's book alone will generally not liquidate you; a market-wide move will, even if the local book never printed the level.

Third, the two-stage exit: liquidation first goes to the order book as a market order, and whatever collateral survives is returned to you. But if equity falls below two-thirds of maintenance margin first, the backstop vault takes the position over — and your remaining maintenance margin is not returned. Being liquidated near the line costs the loss; being liquidated through it costs the loss plus your margin.

The design rule that follows: your stop must always be positioned to fire long before mark price can reach the maintenance line. The complete math, worked examples, and buffer rules are in the liquidation guide.

Perps stay pinned to spot through funding — periodic payments between longs and shorts, set by how far the perp trades from the oracle price. On Hyperliquid, funding settles every hour (at one-eighth of the computed 8-hour rate), flows peer-to-peer with no fee taken, and is calculated on the oracle price. When the perp trades rich, longs pay shorts; when it trades cheap, shorts pay longs.

Two consequences for risk management.

As a cost: funding comes out of the same equity that holds your position up. A crowded direction paying persistent funding erodes your buffer while you wait for the thesis — a position that is 'right eventually' can still bleed enough carry to stop you out of margin before eventually arrives. Price the carry before entering anything you intend to hold for days.

As a signal: extreme funding is the market confessing its positioning. When longs are paying heavily, the crowd is long, levered, and fragile — precisely the conditions where squeezes originate. Persistent extreme funding against your intended direction is information; persistent extreme funding in your direction is a warning that you are the crowd.

Every rule above caps what a trade can cost — but only the exit order enforces it. On a venue that trades while you sleep, an exit that lives in your head protects nothing: without a resting stop, your real stop is your liquidation price, on the worst terms available.

Hyperliquid's TP/SL orders trigger on mark price and come in two flavors: market (fires with a 10% slippage tolerance — nearly always exits, at what the book offers) and limit (you pin the price, accepting the risk of no fill in a fast move). For stops, whose job is survival, guaranteed-exit is usually worth the slippage. Exits can be attached to the position — covering its full size — or to the entry order itself as a bracket, so the protection exists from the first second of the fill.

The placement rules are the same ones every surviving trader converges on: stops at structural invalidation rather than pain tolerance, targets at real structure rather than round numbers, a minimum risk-reward before entry, and stops that move in one direction only. The mechanics, the failure modes (gap-throughs, unfilled limit stops, stops behind the liquidation price), and the bracket semantics are covered in the stop loss and take profit guide.

Per-trade discipline can still be defeated at the account level. Four rules close the remaining holes:

• Cap total open risk. If each position risks 1%, five open positions risk 5% — a bad day away from a meaningful drawdown. Cap the sum (3-4% is a common ceiling) and treat the cap as a queue: a new trade waits until an old one closes or moves to break-even.

• Count correlation honestly. A BTC long and an ETH long are not two ideas; most days they are one idea expressed twice. Hyperliquid makes holding both effortless — your risk ledger should merge them into one directional bet and size accordingly.

• Install circuit breakers. A daily loss limit (say 3%) and a losing-streak cooldown (three consecutive stops → reduced size or a forced pause) are not motivational posters; they are the specific counter to revenge trading, which is how ordinary drawdowns become terminal ones.

• Keep the ledger. Every trade recorded with its intent, size, levels, and outcome. Not because journaling is virtuous, but because the rules above only bind if breaking them is visible afterwards.

None of this limits your upside on any single good trade. It limits how much a bad week is allowed to compound.

Here is the honest summary of everything above: you already knew most of it. Nearly everyone does. The catastrophic losses on perp venues are not knowledge failures — they are enforcement failures. The rules exist, and then conviction, tilt, or a 3am chart cancels them exactly once, and once is enough.

That is the case for moving enforcement out of willpower and into machinery. Hyperhelm runs this entire guide as code: every trade intent passes a pre-trade gate that sizes the position from current regime and volatility, caps leverage, checks the stop geometry against the liquidation math, and rejects or resizes what fails — before you sign. Exits go in as native reduce-only brackets at entry. The rules fire on every trade, including the ones you feel strongest about, because the gate does not know what conviction is.

You keep the override — it is your wallet and your call — but overrides are logged, and the public benchmark scores governed versus ungoverned execution from the same intents. That is risk management with a receipt: not a promise of discipline, but a measured record of what the discipline added. The cockpit is free to open, and the gate's verdict is visible before you ever approve anything.