Crowded trades and cascading liquidations: the hidden risk in copy trading

What makes a trade crowded

Crowding is not about how many people like an idea. It is about how much capital holds the same exposure, at what prices, with how much leverage — because those three things determine where the forced sellers are.

A large position held by patient, unlevered capital is not crowded in the sense that matters. A moderate position held by many leveraged accounts that all entered within the same range is, because a modest adverse move converts a lot of it into forced supply simultaneously.

Similar entries, similar leverage, similar liquidation prices

Liquidation happens when maintenance margin is breached, and where that occurs is a function of entry price, leverage and remaining margin.

If a thousand accounts enter within a narrow price band at comparable leverage, their liquidation levels fall within a narrow band too. That band is not distributed across the price range — it is a wall.

Markets do not treat that wall as neutral. Concentrated liquidation levels are a known feature of leveraged venues, and price reaching them produces a burst of forced market orders that no one is choosing to send.

• Crowding is measured in clustered liquidation levels, not in sentiment.
• Leverage determines how close the cluster sits to the current price.
• Simultaneity is the danger: forced flow arriving all at once, not in sequence.

Copy trading manufactures crowding by design

Ordinary crowding emerges when independent traders reach similar conclusions. Copy crowding is engineered: one leader's decision is replicated mechanically across every follower account, within seconds, in the same market and direction.

That produces tighter clustering than organic crowding does. Independent traders enter at different times and sizes; followers of the same leader enter within a narrow window at proportional sizes.

The effect scales with the leader's popularity rather than with their skill, and the two are not the same. A leader whose visibility has grown faster than their strategy's capacity accumulates followers whose combined position is larger than the market the strategy was built for.

There is a second-order version too. When multiple systems select leaders using similar metrics — recent PnL, ROI, win rate — they tend to select overlapping sets of leaders. Independent selection processes converging on the same wallets reproduces the crowding at the level of the whole market.

How crowding amplifies a liquidation cascade

A cascade is a feedback loop with three steps: price reaches a cluster of liquidation levels, forced closes execute as market orders, and that flow moves price into the next cluster.

Nothing about the loop requires panic, news, or irrational behaviour. It is a mechanical consequence of leverage distribution and order book depth.

Forced selling into a book that just thinned

Liquidations are the worst possible orders arriving at the worst possible moment. They are unconditional — the liquidation engine takes whatever price the book offers — and they arrive precisely when market makers have widened or pulled quotes because volatility spiked.

So the same event that generates the forced flow also removes the liquidity that would have absorbed it. Depth is not a constant; it is thinnest exactly when it is needed most.

This is why realised slippage during a cascade bears no relation to slippage in normal conditions, and why any cost estimate calibrated on calm markets understates what a stress event costs.

Why followers are liquidated before the leader

This is the asymmetry that matters most and is least discussed.

A follower enters after the leader, which in a moving market usually means a worse price. A worse entry at the same leverage places the liquidation level closer to the current price. The follower is therefore in the earlier tier of the cascade.

The leader may also have unequal margin flexibility. A leader with substantial free collateral can absorb an adverse excursion that liquidates a follower running the same nominal leverage on a fully deployed account.

The result is a scenario that looks like bad luck and is actually structural: the leader survives a wick, closes the trade profitably later, and reports a winning trade — while followers were liquidated during the excursion and never saw the recovery. The leader's track record is honest. It is simply not a description of what followers experienced.

• Later entry means a liquidation level closer to spot.
• Free collateral is not mirrored, so margin resilience differs.
• A leader's winning trade and a follower's liquidation are compatible facts.

Single-trader versus multi-trader systems under crowding

Copying one leader means your entire account sits in whatever crowd that leader is part of. If the position is crowded and the cascade comes, there is nothing else in the account to be unaffected.

A multi-leader basket helps, but only to the degree the leaders are genuinely uncorrelated. Diversification across ten leaders who are all long the same major perp is diversification of names, not of exposure — a cascade in that market hits every sleeve at once.

The useful distinction is between leader count and exposure overlap. Style diversity, market diversity and horizon diversity are what actually reduce cascade exposure; roster size on its own does not.

This is also why selection criteria that emphasise recent returns tend to reduce real diversification over time. In a strongly trending market, the traders who score best are frequently the ones positioned the same way.

What sub-account isolation does and does not contain

Running one Hyperliquid sub-account per mirrored leader changes the blast radius of a cascade without changing its likelihood.

What it contains: a liquidation in one sleeve consumes that sleeve's margin and stops there. The other sleeves keep their positions, their margin and their liquidation levels, because they were never sharing a margin pool. In a single cross-margined account, a liquidation in one leader's position draws down collateral supporting every other position and can trigger further liquidations that had nothing to do with the original trade.

What it does not contain: correlated exposure. If several sleeves hold the same directional exposure in the same market, isolation means each is liquidated separately rather than together. The losses are the same; they are simply attributed accurately.

Isolation is a containment mechanism, not a hedge. It bounds contagion inside your account. It does nothing about crowding in the market, and it does nothing about several sleeves being wrong at once.

Practical ways to reduce crowding risk

None of these eliminate the risk. They reduce how much of it lands in one place.

Cap aggregate exposure per market across all sleeves, not just per sleeve. Per-leader limits do not stop five leaders from adding up to a single concentrated position.

Cap mirrored leverage independently of the leader's own leverage. Leverage is the single largest determinant of how close your liquidation level sits to the current price, and it is the one parameter a follower can control without altering the strategy's direction.

Prefer style diversity over leader count. Two leaders with genuinely different horizons and markets reduce cascade exposure more than six leaders running the same playbook.

Be sceptical of capacity. A strategy that works in a deep major perp may not survive being replicated at scale in a thin alt perp, and thin books are where cascades are most violent.

Treat popularity as a risk factor rather than as validation. The more capital following a leader, the more crowded their positions are by construction.

• Aggregate exposure limits per market, across all sub-accounts.
• Independent leverage caps that ignore what the leader uses.
• Style, market and horizon diversity over roster size.
• Extra caution in thin markets where depth disappears fastest.

Honest limitations

Crowding cannot be measured precisely from outside. Aggregate positioning, the distribution of liquidation levels and the true depth of a book under stress are all partially observable at best, and estimates of them are worst in exactly the conditions where they matter.

Diversification and isolation reduce the damage a single cascade does to an account. They do not prevent a broad market event from affecting most positions simultaneously, and correlation between strategies rises sharply in stress — the diversification you measured in calm conditions is not the diversification you have during a cascade.

Risk controls also act on a delay. A notional cap prevents a position from becoming too large; it does not close a position mid-cascade at a price you would have chosen.

Trading perpetual futures carries a substantial risk of loss, including the loss of your entire position. Cascades are one of the mechanisms by which that loss is realised quickly.

Conclusion

Crowding is the cost of copying something popular. It is inherent to the mechanism rather than a flaw in any particular implementation, and it explains a specific failure that surprises followers: being liquidated on a trade the leader ultimately won.

The structural responses are unglamorous — leverage caps, aggregate exposure limits, genuine style diversity and per-leader isolation so one cascade does not become an account-wide event. If you want the follower-side margin mechanics in detail, the liquidation risk note covers them, and the risk controls note covers what sits on top of diversification.