Sub-account isolation: why netting destroys your portfolio

What netting actually is

On Hyperliquid, as on every perpetual futures venue that uses one-way positions, an account holds a single net position per market. You do not hold a long ETH position and a short ETH position side by side. You hold one ETH position, and every order you send moves that one number.

Buying when you are short does not open a new trade. It reduces or closes the short. Selling when you are long does not open a hedge. It reduces or closes the long. The matching engine does not know or care that two different strategies produced those two orders — it only sees the account, the market, and the resulting net size.

For a single trader running a single strategy, this is convenient and correct. For a copy trading system mirroring several independent leaders into one account, it is the mechanism by which the portfolio destroys itself. Two leaders you selected precisely because their approaches differ will, sooner or later, disagree in the same market. When they do, the account resolves that disagreement by cancelling both of them.

A concrete example: two leaders, one ETH position

Suppose you allocate to two leaders. Leader A runs trend continuation. Leader B runs mean reversion. That is a reasonable pairing: their edges are supposed to be uncorrelated, and a diversified basket is exactly where uncorrelated behaviour is valuable.

ETH breaks out. Leader A goes long ETH with 5x leverage on a meaningful share of their book. Forty minutes later, ETH extends, and Leader B — reading the same move as overextended — goes short ETH with similar size. Both leaders are now in a position, in their own accounts, on their own terms. Both are doing exactly what you copied them for.

Inside your single copy account, the sequence is different. Your mirror of Leader A opens a long ETH position. Your mirror of Leader B then sends a sell order of comparable size. That order does not open a short. It closes the long. Your account is now flat ETH, or holds a small residual in whichever direction happened to be larger.

Consider what you now own. Two leaders are, on their own screens, carrying live ETH risk. Your account carries none. If Leader A is right and the trend continues, you make nothing from it, because the position was closed before the move. If Leader B is right and price reverts, you make nothing from that either. Both leaders can be correct in sequence and your ETH sleeve can still return zero — minus costs.

Then the exits arrive. Leader A closes their long: your account, being flat, receives a sell order that now opens an unintended short. Leader B closes their short: a buy order that closes it again. You have executed four round trips of ETH exposure without ever holding the exposure either leader intended, and every one of those fills was a decision no one in the system actually made.

• Neither leader's thesis is expressed in your account.
• Order sequencing, not strategy, determines what you hold.
• Exits from a netted state open positions instead of closing them.

The real costs of netting

Netting is usually described as a technical inconvenience. It is more accurate to treat it as four separate, compounding costs — three of which are invisible on a PnL screen.

Lost exposure

This is the direct cost. You allocated capital to two independent bets and ended up holding neither. The diversification you paid for was not diluted; it was deleted in that market for the duration of the overlap.

Worse, the loss is asymmetric in an unhelpful direction. Netting removes exposure precisely when two leaders disagree — which is when the market is most contested and when the winning side's move tends to be largest. The positions that survive netting are the ones where your leaders agreed, meaning your effective portfolio drifts toward correlated, consensus trades. That is the opposite of the intended construction.

Double fees and funding leakage

Every one of those offsetting fills is a taker or maker fee, plus slippage on entry and exit. In the example above you paid the round-trip cost of two ETH positions and held, on net, close to nothing. The fees are real; the exposure was not.

Funding is the second leak. A netted or near-flat position pays or receives almost no funding, so leaders whose edge is partly a funding carry lose that component entirely. But during the windows when the netting is imperfect — an order arrives late, sizes differ, one leader scales in — you do pay or receive funding on a residual position that no leader chose to hold. You are exposed to the cost structure of a position without the intent behind it.

In a high-frequency basket, this is not a rounding error. It is a continuous drag proportional to how often your leaders overlap, and it scales with the number of leaders you add — which is to say, it gets worse exactly as you try to diversify further.

Distorted performance tracking

Once orders from multiple leaders hit one account, fills stop being attributable. Your account shows an ETH position of some size at some average entry. It does not show that 60% of it came from Leader A, that Leader B's short was absorbed into the average, or that the realised PnL on the close belongs partly to a position neither of them was running.

That breaks the feedback loop that any serious allocation system depends on. You cannot score a leader on their contribution to your account if your account never expressed their positions. You cannot decide to replace a leader for underperformance when the underperformance may be a netting artefact. And you cannot tell the difference between a leader whose edge decayed and a leader whose trades were repeatedly cancelled out by someone else's.

Public leaderboard statistics do not solve this. They describe what the leader's own account did, which is exactly the thing your netted account failed to reproduce. The gap between the two is unmeasurable from inside a merged account.

Broken risk attribution

The most serious cost is on the risk side. In a single account, margin is shared. Every leader's position consumes the same margin pool and moves the same liquidation price. There is no such thing as one leader's drawdown.

The practical consequences are concrete. A leader with disciplined sizing can be liquidated out of their position because a different leader loaded up on an unrelated alt perp and drained account margin. Per-leader risk limits become unenforceable, because a limit expressed as a share of a shared margin pool depends on what everyone else is doing. Setting leverage is ambiguous: leverage is an account-level property in a netted book, so two leaders running 3x and 10x cannot both be represented faithfully.

And removing a leader becomes a portfolio-wide event. To stop copying Leader B, you must unwind their share of a merged position — which means sending orders that also change the exposure of every other leader who traded that market.

• Shared margin means one leader can liquidate another.
• Per-leader risk limits cannot be enforced against a common pool.
• Per-leader leverage settings cannot be represented at all.
• Replacing one leader disturbs every leader in the same market.

Why multi-leader copying is almost impossible without isolation

Systems that try to run several leaders in one account end up building a netting reconciliation layer: tracking notional exposure per leader in an internal ledger, computing a target net position for the account, and sending the difference. It is a legitimate engineering approach and it fails in predictable ways.

The internal ledger is a claim, not a fact. It says Leader A owns +2 ETH and Leader B owns -2 ETH while the venue holds zero. Any partial fill, rejected order, latency spike, or leader action between snapshots desynchronises the ledger from the actual account, and the drift is silent. Reconciliation also cannot manufacture exposure that was never held: if the account is flat, the profit from Leader A's correct call does not exist to be attributed, no matter how the ledger apportions it.

Liquidation is where the abstraction breaks completely. Liquidation happens at the account level against real margin, not against your ledger. A synthetic per-leader accounting layer can tell you who you think should have been liquidated. It cannot stop the venue from closing a position that belonged to someone else's strategy.

How sub-account isolation solves it

Isolation removes the problem at its source rather than compensating for it. Each leader is mirrored into its own sub-account, funded with its allocated share of capital. Because netting is an account-level property, giving each leader a separate account means there is nothing to net.

Leader A's long ETH lives in Leader A's sub-account. Leader B's short ETH lives in Leader B's sub-account. Both positions exist simultaneously, at the venue, with real margin behind them. If both are eventually right, both contribute. If both are wrong, both losses are attributed to the leader who produced them.

Everything downstream becomes well-defined. Margin is per sub-account, so a blow-up is bounded by that leader's allocation and cannot reach the others. Liquidation price is per sub-account and computed only from that leader's own positions. PnL is per sub-account, which means scoring and replacement decisions run on measured contribution rather than inference. Leverage and risk limits can be set per leader, because each leader has an account of their own to apply them to. And removing a leader is a local operation: close that sub-account's positions, reclaim the capital, leave every other position untouched.

The trade-off is honest and worth stating: isolation requires splitting capital across accounts, and each sub-account needs enough margin to hold a meaningful position without sitting close to liquidation. That is the reason full diversification across the whole basket has a capital threshold, and why smaller accounts start with a reduced number of isolated leaders rather than a merged account holding all of them. Merging would look like more diversification while delivering less.

• No netting, because there is no shared position to net into.
• Bounded blast radius: one leader's failure is capped at their allocation.
• Measurable per-leader PnL, so scoring reflects real contribution.
• Per-leader leverage and risk limits are actually enforceable.
• Leader replacement without touching unrelated exposure.

Conclusion

Netting sounds like an implementation detail and behaves like a portfolio design constraint. It determines whether a multi-leader basket is genuinely diversified or merely appears to be, whether performance data means anything, and whether risk can be attributed to the decisions that caused it.

That is why isolation is not a feature added on top of a copy trading engine. It is the precondition for one. Any system that mirrors more than one leader into a single perpetual futures account is running a portfolio whose composition is decided by order timing rather than by allocation.

Isolation does not make copy trading safe. Perpetual futures remain leveraged instruments and an isolated sub-account can still be liquidated in full. What it does is ensure that what you hold is what you chose to hold — which is the minimum requirement for evaluating anything else. If you are assessing a diversified copy trading system, the account structure is the first thing worth asking about.