Published
The short version
When a leveraged position gets liquidated, the exchange tries to close it at the bankruptcy price β the price at which the trader's margin hits exactly zero. If the market is moving fast, the actual fill often happens at a worse price than that. Someone has to eat the difference. That's what the insurance fund is for.
If the fund is healthy, it quietly absorbs the shortfall and nobody else notices. If a cascade of liquidations drains it faster than it can refill, the exchange has one remaining option: force-close profitable traders on the other side of the trade, at the losing trader's bankruptcy price, to keep the system solvent. That's auto-deleveraging (ADL) β and it's the part most traders never read about until it happens to them.
Where the fund's money actually comes from
The insurance fund isn't exchange profit sitting in a vault. It's built almost entirely from liquidations themselves:
- Liquidations that close better than bankruptcy price. If your position is liquidated and the forced close fills above your bankruptcy price (for a long), the leftover margin β after fees β doesn't go back to you. It goes into the insurance fund.
- Liquidations that close worse than bankruptcy price. These do the opposite β they draw money out of the fund to cover the deficit.
- Initial exchange seeding. Most exchanges seed the fund at launch and top it up during low-volatility periods when liquidations are mild and net-positive for the fund.
In calm markets, the fund grows β most liquidations close at a small profit to the fund. In a fast, one-directional crash (think a 15% BTC wick in minutes), liquidation engines can't fill every forced order at a clean price, and the fund can shrink fast.
How big are these funds, really?
Public dashboards give a rough sense of scale. In early 2026, Binance's futures insurance fund balances have shown up in the roughly $500Mβ$800M range (separate from Binance's $1B SAFU user-protection fund, which covers a different kind of risk β exchange hacks, not trading losses). Bybit's USDT perpetual insurance fund has held above $400M. These numbers move daily depending on market conditions and are published live on most major exchanges' insurance fund pages.
The order of operations when you get liquidated
- Your margin hits the maintenance margin threshold. The liquidation engine takes over your position.
- The engine tries to close it on the open market.
- If the fill is better than your bankruptcy price β the leftover goes to the insurance fund.
- If the fill is worse than your bankruptcy price β the insurance fund covers the gap. You still only lose your margin (isolated mode); the fund absorbs the rest.
- If the insurance fund itself can't cover the gap for that trading pair β auto-deleveraging activates.
Step 5 is the one most traders have never heard of until it happens.
Auto-deleveraging (ADL): the mechanism of last resort
ADL force-closes positions on the winning side of the trade β people who did nothing wrong, aren't anywhere near liquidation themselves, and may not even know it's coming β to absorb the loss the insurance fund can't cover. It's the alternative to socializing the loss across every user on the platform (an older, even less popular model some exchanges used before insurance funds and ADL became standard).
ADL doesn't pick targets randomly. Exchanges rank opposing positions by a formula designed to deleverage the riskiest, most profitable positions first:
| Exchange | How ADL ranks who gets closed first |
|---|---|
| Binance | PnL% Γ effective leverage (when in profit) β the most profitable, most leveraged positions rank highest and get force-closed first |
| Bybit | Triggers when a single pair's 8-hour drawdown reaches the insurance fund's 8H PnL drawdown trigger line for that period |
If you're deleveraged, your position closes at the liquidated trader's bankruptcy price β not the current market price, and not a price you chose. You keep the realized PnL up to that point, but the position itself is gone, whether you wanted out or not.
Why this matters even if you never get liquidated
You can run a perfectly disciplined trade β modest leverage, a stop-loss that never gets touched, a position sitting comfortably in profit β and still have it closed without warning because someone else, on the other side of the order book, got liquidated into a thin market. ADL isn't a punishment for bad risk management; it's a structural risk of trading any leveraged perpetual product, full stop.
Most exchanges show a live ADL risk indicator (often a tiered light or bar) on your position page once you're in meaningful profit. If you're holding a large, highly leveraged winning position, that indicator is worth checking β it's the only early warning you'll get.
What you can actually do about it
- Take partial profits on big winners. ADL ranks by PnL% Γ leverage β reducing size on a position that's run up a lot moves you down the queue.
- Lower leverage on positions you plan to hold through volatility. Same logic β effective leverage is one of the two inputs in the ranking.
- Prefer liquid pairs for large size. BTC and ETH perps have deeper insurance fund allocations and liquidation engines that can usually fill without triggering ADL. Thin altcoin perps are the pairs where ADL actually fires in practice.
- Understand it's not the same risk as your own liquidation. Managing your own liquidation price (see the liquidation calculator) protects you from your own leverage. It does nothing about ADL risk, which comes from the other side of the book.
Related calculators
If this made you think about your own liquidation distance, these are the tools to check it:
- Liquidation price calculator β any leverage, any coin, long or short
- Leverage table β see all leverages at once
- Bybit liquidation calculator β Bybit-specific maintenance margin rates
- Binance liquidation calculator β Binance-specific rates
- Position size calculator β size trades so a stop-loss costs a fixed, known % of your account
- RektCalc Learn hub β liquidation, funding rates, margin modes and more, explained plainly