Hyperliquid Position Sizing: The Formula, the Mistakes, and How to Automate It

Ask a struggling perp trader how they size and the answer is a leverage number: 'I trade 10x.' That answer contains no information about risk. 10x of what fraction of the account? With the stop how far away? A 10x position with a 1% stop and a 3x position with a 15% stop are wildly different risks wearing similar clothes.

Leverage answers a financing question — how much margin finances the notional. Position sizing answers the only question that decides survival: how much do I lose if this trade is wrong? Confuse the two and your risk per trade becomes a random variable that swings with your stop placement, your mood, and whatever leverage felt appropriate that day. No edge survives random bet sizing.

The fix is one formula, applied without exception. It takes thirty seconds per trade, and it is the single highest-value habit in leveraged trading.

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

The three inputs, in the order you should decide them:

• Account equity — what is actually in the account. Not what will be there after this trade works.

• Risk fraction — the percentage of equity this trade may cost. 0.5% is conservative, 1% is standard, 2% is aggressive. Above that you are not sizing positions, you are scheduling drawdowns: ten consecutive losses at 5% each is a 40% hole, which needs a +67% run just to get back to even.

• Stop distance — the percentage between entry and the level where the idea is invalid. This comes from structure — below the swing low, beyond the range boundary — never from what loss feels comfortable.

Worked once: $10,000 account, 1% risk, BTC setup with invalidation 4% below entry. Size = ($10,000 × 0.01) ÷ 0.04 = $2,500 notional. If the stop hits, you lose $100 — exactly the 1% you chose. Tighten the stop to 1% (a well-defined level nearby) and the same $100 of risk supports $10,000 notional. The stop's geometry sets the size; the risk never changes.

Leverage appears nowhere above. It falls out at the end: finance the $2,500 with $2,500 margin (1x) or $500 (5x) — the risk is identical. What changes is liquidation distance, which is a constraint to check, not a dial to crank.

Hyperliquid's margin system gives the formula sharp edges. BTC's maintenance margin is 1.25% of notional (half the initial margin at its 40x max leverage), so an isolated position's liquidation sits roughly at (margin fraction − 1.25%) adverse.

Example 1 — the formula, then the financing. $10,000 account, 1% risk, stop 4% away → $2,500 notional. Allocate $500 isolated margin (5x): liquidation sits ~18.8% away, nearly five times the stop distance. The stop will always fire first — the liquidation engine is irrelevant to this trade, which is exactly the goal.

Example 2 — the leverage-first trader, for contrast. Same $10,000 account. 'BTC allows 40x, I'll be conservative and use 20x': $10,000 margin × 20 = $200,000 notional. The same 4% stop now represents an $8,000 loss — 80% of the account on one trade. And the liquidation sits ~3.8% away: closer than the stop, so the stop never even gets its turn. Every number is legal on the venue; the account is simply pre-committed to dying on the first normal fluctuation.

Example 3 — the quick reference. At 1% risk, notional as a multiple of equity is just 1% ÷ stop: a 1% stop → 1.0× equity, 2% → 0.5×, 4% → 0.25×, 8% → 0.125×. If a 'good setup' needs more size than the formula allows, the setup is fine — your risk budget is the binding constraint, as designed.

• Sizing by conviction. 'I'm really sure about this one' is how the largest position of the month meets the drawdown of the year. Conviction is already expressed in taking the trade at all; it must not also set the size — confidence peaks are exactly where regimes turn.

• Equal dollars on every trade. Same notional regardless of stop distance means your actual risk swings randomly with stop geometry — a 2% stop risks half of what an identical-size 4% stop risks. The formula exists to make risk constant, not notional.

• Risking a fraction of margin instead of equity. '1% of my margin' on a 10x position is 1% of one-tenth of your money — an accounting trick that hides 10x the intended risk. The risk fraction always applies to account equity.

• Doubling after losses. Martingale sizing converts a routine losing streak into a terminal event. If anything, size should shrink during streaks — which is what a loss-based cooldown enforces mechanically.

• Counting correlated trades as separate risks. A BTC long and an ETH long are, most days, one directional bet expressed twice. If each risks 1%, the book risks ~2% on one idea. Merge correlated exposure in the risk ledger and size the idea, not the tickets.

• Stops inside the noise. A stop tighter than the market's normal fluctuation gets tagged constantly — so the formula hands you a big position that stops out on randomness. The fix is never to widen the stop to 'give it room' at the same size: widen the stop because structure demands it, and let the formula shrink the size accordingly.

A subtle strength of stop-based sizing: it is volatility-aware for free, as long as your stops are honest.

In a quiet regime, structural invalidation sits close — stops are tight, and the formula permits size. When volatility expands, the same structural logic pushes invalidation further away — stops widen, and the formula cuts size automatically. Risk in dollars stays constant across regimes while notional breathes with the market. Traders who fix their stop percentage regardless of conditions are implicitly betting that volatility never changes; the market settles that bet a few times a year, always in the same direction.

The discipline this requires is placing stops from structure (or a volatility measure like ATR) rather than from a fixed habit — and then accepting the smaller size in wild conditions instead of overriding it. The override is always tempting, because expanding volatility usually arrives with the most exciting-looking setups. That correlation is not a coincidence; it is the trap. High-volatility regimes are precisely where oversized positions meet gap-throughs and the liquidation math from our liquidation guide stops being theoretical.

Nothing in this article is hard. That is exactly why the failures are so telling: the formula takes thirty seconds, and it still gets skipped — at the top of a winning streak, after a frustrating stop-out, in the middle of a breakout candle. Arithmetic loses to adrenaline with remarkable consistency.

So make the arithmetic non-optional. In Hyperhelm, position size is not something you type — it is computed by the risk governor for every trade intent: the current regime, volatility, and data quality set the maximum size fraction and leverage cap; the stop comes from the structure map, not from hope; and the intent arrives fully sized, with the risk-reward stated. If the geometry fails — size too big for the stop, liquidation too close, risk budget already spent — the gate resizes or rejects it before the order exists.

You can override the governor; the override is logged, and the public benchmark scores what governed and ungoverned execution would each have done from the same intents. Over months, that produces the honest version of the claim every trader makes about their own discipline: not 'I usually size correctly,' but a measured record. The complete risk-management guide covers the rest of the system this plugs into.