Liquidation Cascades - Chain Reactions in Crypto Markets
TL;DR
- Liquidation cascade = A chain reaction where forced selling pushes prices down, triggering more liquidations, which causes more forced selling
- Why crypto is vulnerable: High leverage (up to 100x), fragmented liquidity across dozens of exchanges, 24/7 markets with no circuit breakers, thin order books, and highly correlated assets
- Historical examples: March 2020 ("Black Thursday") saw BTC drop 50% in 24 hours with $1B+ liquidated on BitMEX alone. May 2021 saw $8B liquidated across crypto in 24 hours. FTX/Alameda in November 2022 triggered cascading liquidations across the entire industry
- The doom loop: Price drops 5% → overleveraged positions get liquidated → liquidation sell orders hit thin books → price drops another 5% → more positions hit liquidation → repeat until leverage is flushed out
- Exchange defenses: Tiered liquidation, backstop liquidity providers, ADL, insurance funds, circuit breakers, liquidation throttling
- Backpack's approach: 3-tier liquidation system (on-book → backstop → ADL) with probabilistic throttling (50% probability, 10% per loop) that spreads liquidation pressure over ~20 seconds instead of dumping everything at once
1. What Is a Liquidation Cascade?
A liquidation cascade is a self-reinforcing feedback loop where forced position closures drive prices further in the direction that triggers additional forced closures. It is the market equivalent of a stampede — one person runs for the exit, which causes others to run, which causes more people to run, until the entire crowd is crushing itself against the door.
In leveraged trading, every position has a liquidation price — the price at which the exchange forcefully closes the position to prevent the trader's losses from exceeding their collateral. When the market moves sharply and hits one trader's liquidation price, the exchange sells their position. That selling pushes the price down further. If it pushes far enough, it hits the next trader's liquidation price. Their position gets sold too. More price impact. More liquidations. The loop continues until either:
- The price stabilizes (enough liquidity absorbs the selling), or
- Most leveraged positions in that direction have been wiped out
The danger is that this process is nonlinear. A 5% price drop might liquidate $10M in positions. Those $10M in forced sells might cause another 3% drop. That 3% drop might liquidate $50M more. The second wave is larger because more positions cluster near the current price. Each wave can be bigger than the last.
The Doom Loop:
Price Drop
│
▼
┌───────────────┐
│ Positions │
│ Hit MMF │──────────────────────┐
└───────────────┘ │
│ │
▼ │
┌───────────────┐ │
│ Forced │ │
│ Selling │ │
└───────────────┘ │
│ │
▼ │
┌───────────────┐ │
│ Order Book │ │
│ Absorbs │ │
│ (with impact)│ │
└───────────────┘ │
│ │
▼ │
┌───────────────┐ │
│ Price Drops │──────────────────────┘
│ Further │ Feedback loop repeats
└───────────────┘ until leverage is flushed2. The Mechanics Step by Step
Phase 1: The Initial Trigger
Every cascade starts with an exogenous shock — something that moves the price sharply before the feedback loop begins. Common triggers include:
- A large spot sell (whale dumping, fund redemptions, token unlock)
- Negative news (regulatory action, hack, insolvency)
- Contagion from another asset or exchange
- A sudden drop in stablecoin peg confidence
The initial move itself is not the cascade. The cascade is what happens next.
Phase 2: First Wave of Liquidations
The initial price drop pushes some accounts below their maintenance margin fraction (MMF). These are typically the most leveraged positions — traders using 50x or 100x leverage, whose liquidation prices are only a few percent from their entry.
The exchange begins force-closing these positions. For long positions, this means selling into the market. For short positions during a price spike, it means buying. Either way, the liquidation orders consume resting liquidity on the opposite side of the book.
Phase 3: Liquidity Consumption
Liquidation orders are typically market orders or aggressive limit orders (IOC crossing the spread). They eat through the order book from the top. In a thin market:
Before liquidation: After $500K liquidation sell:
BIDS: BIDS:
$100.00 50 BTC ← best bid (consumed)
$99.50 30 BTC (consumed)
$99.00 20 BTC $99.00 8 BTC ← new best bid
$98.50 40 BTC $98.50 40 BTC
$98.00 60 BTC $98.00 60 BTC
Spread: $0.10 Spread: $1.10
Price impact: $0 → $1.00+The key insight: liquidity is not uniformly distributed. Order books are thinnest near the current price and deeper further away. But during a cascade, resting orders get cancelled as market makers pull their quotes, making the book even thinner precisely when it needs to be deep.
Phase 4: Second Wave (The Amplification)
The price impact from Phase 3 pushes more accounts below their MMF. This second wave is often larger than the first because:
- Clustering: Many traders use similar leverage and entry prices, so their liquidation prices cluster
- Cross-margin effects: A drop in one position reduces the account's total equity, which can push other positions below MMF
- Collateral devaluation: If traders are using the same asset as collateral (e.g., BTC to margin BTC-perp), the collateral loses value at the same time as the position
Phase 5: Market Maker Retreat
Professional market makers have risk limits. When volatility spikes and order flow becomes toxic (mostly liquidation-driven), they:
- Cancel resting orders (reducing book depth)
- Widen spreads (increasing trading costs)
- Reduce position limits (posting less size)
- In extreme cases, disconnect entirely
This removes the shock absorber from the market exactly when it is needed most. The book thins out, which means each subsequent liquidation has more price impact, which triggers more liquidations.
Phase 6: Cross-Exchange Contagion
Crypto assets trade on dozens of exchanges simultaneously. A cascade on one exchange drives the price down there, which:
- Moves the index price that other exchanges use for mark price calculation
- Creates arbitrage opportunities where bots sell on other exchanges to align prices
- Triggers liquidations on those exchanges, which feed back into step 1
This cross-venue feedback can turn a single-exchange event into a market-wide crisis.
Phase 7: Resolution
The cascade ends when one or more of these occurs:
- Leverage flush: Most high-leverage positions have been liquidated; remaining positions have lower leverage and wider distance to liquidation
- Buyer absorption: Enough buyers step in (bargain hunters, arbitrageurs, long-term holders) to absorb the selling pressure
- Exchange intervention: Circuit breakers, trading halts, or liquidation throttling slow the process enough for the book to rebuild
- Backstop liquidity: Designated market makers or insurance funds absorb the remaining liquidations off-book
3. Historical Examples
March 12, 2020 — "Black Thursday"
What happened: COVID-19 panic triggered a global sell-off across all asset classes. BTC dropped from ~$7,900 to ~$3,800 in 24 hours — a 50%+ decline.
The cascade mechanics:
- Global equity markets crashed, triggering margin calls across traditional finance
- Traders sold crypto to cover margin calls in other markets (contagion from TradFi)
- BitMEX, the dominant leveraged trading venue at the time, saw over $1 billion in long liquidations in a single day
- BitMEX's order book was overwhelmed — liquidation orders couldn't fill because there were no bids
- The price on BitMEX briefly hit $3,596 while other exchanges were trading $1,000+ higher
- BitMEX's insurance fund was depleted; ADL kicked in for the first time at scale
- BitMEX eventually went offline for "maintenance" (widely interpreted as a circuit breaker), which ironically helped stabilize the market by stopping the liquidation engine
Key lesson: A single exchange's liquidation engine, running unchecked with no throttling and insufficient backstop liquidity, can drive prices far below fair value. BitMEX's downtime was accidental proof that pausing liquidations can prevent cascades.
May 19, 2021 — The $8 Billion Day
What happened: China announced a renewed crackdown on crypto mining and trading. BTC dropped from ~$43,000 to ~$30,000 in hours. ETH fell from ~$3,400 to ~$1,800.
The cascade mechanics:
- The news triggered spot selling, which pushed prices down ~10%
- That 10% drop was enough to start liquidating the massive leverage that had built up during the 2021 bull run
- Over $8 billion in leveraged positions were liquidated across all exchanges in 24 hours
- Funding rates had been deeply positive (longs paying shorts), indicating extreme long-side crowding
- As longs were liquidated, the selling drove prices down further, creating a cascading effect across BTC, ETH, and altcoins simultaneously
- Multiple exchanges experienced outages due to traffic spikes, which trapped traders who couldn't add margin or close positions manually
Key lesson: Correlated positioning (everyone long, everyone in the same assets) means cascades affect the entire market at once. When funding rates are extreme, the system is loaded with potential energy.
November 2022 — FTX/Alameda Collapse
What happened: Alameda Research's balance sheet was revealed to be heavily concentrated in FTT (FTX's exchange token). Binance announced it would sell its FTT holdings. This triggered a bank run on FTX, which turned out to be insolvent.
The cascade mechanics:
- FTT price collapsed from ~$22 to ~$1 as confidence evaporated
- Alameda's positions were liquidated across multiple venues, dumping billions in assets
- Assets that Alameda held (SOL, SRM, MAPS, and other Solana ecosystem tokens) crashed as the market priced in forced selling
- FTX froze withdrawals, trapping ~$8 billion in customer assets
- Contagion spread to every firm with FTX/Alameda exposure: BlockFi, Genesis, Voyager (already struggling), and dozens of smaller funds
- The total crypto market cap dropped by ~$200 billion in a week
- BTC fell from ~$21,000 to ~$15,500 — a new cycle low
Key lesson: Cascades are not just about leveraged positions on exchanges. They can be institutional — one large entity's insolvency forces asset sales, which impair other entities' balance sheets, which forces more sales. The FTX cascade was a credit contagion event that played out like a bank run.
4. Why Crypto Is Especially Vulnerable
High Available Leverage
Many crypto exchanges offer 50x to 125x leverage. At 100x leverage, a 1% adverse move wipes out the entire margin. This means liquidation prices are extremely close to entry prices, and even minor volatility can trigger cascading liquidations.
| Leverage | Margin | Liquidation Distance |
|---|---|---|
| 5x | 20% | ~18-20% move |
| 20x | 5% | ~4.5-5% move |
| 50x | 2% | ~1.8-2% move |
| 100x | 1% | ~0.9-1% move |
Fragmented Liquidity
BTC trades on 50+ exchanges simultaneously. Total global order book depth is split across all of them. An exchange with 5% of global volume has only 5% of global liquidity to absorb its liquidations. But it may have 5% of global leverage to liquidate — and that leverage is concentrated among its own users, not distributed across the ecosystem.
24/7 Markets With No Halts
Traditional equity markets have circuit breakers (NYSE Rule 80B triggers a 15-minute halt after a 7% S&P 500 decline). Crypto has no equivalent. Markets trade continuously, including weekends and holidays when liquidity is typically at its lowest. Some of the worst cascades have occurred during off-hours when fewer market makers are active and order books are thinnest.
Thin Order Books
Crypto order books, even for BTC, are thin relative to the notional value of open interest. It is common for total open interest in BTC perpetuals across exchanges to exceed $15-20 billion, while the combined order book depth within 2% of mid-price is only $200-300 million. If 10% of open interest gets liquidated, it represents 5-10x the available liquidity.
Correlated Assets
In a crypto crash, almost everything falls together. BTC, ETH, SOL, and altcoins have high return correlations during stress events (often 0.9+). This means a cascade in one asset impairs collateral for positions in other assets, creating cross-asset cascading. A trader using SOL as collateral for a BTC-perp position gets hit twice: their BTC position loses money and their SOL collateral loses value.
Reflexive Collateral
Many crypto traders use the same asset they are trading as collateral. A BTC long position collateralized with BTC has a reflexive exposure: when BTC drops, both the position and the collateral lose value simultaneously. This creates an accelerating liquidation dynamic compared to cash-margined systems.
5. Anatomy of a Cascade — Numerical Walkthrough
This section walks through a simplified cascade in detail. The numbers are stylized to illustrate the mechanics clearly.
Setup
Market: BTC-PERP, starting price $100 (scaled for simplicity)
Order book depth (bid side):
| Price | Size (BTC) | Cumulative Notional |
|---|---|---|
| $100.00 | 50 | $5,000 |
| $99.00 | 40 | $8,960 |
| $98.00 | 30 | $11,900 |
| $97.00 | 25 | $14,325 |
| $96.00 | 20 | $16,245 |
| $95.00 | 20 | $18,145 |
| $94.00 | 15 | $19,555 |
| $93.00 | 15 | $20,950 |
| $92.00 | 10 | $21,870 |
| $91.00 | 10 | $22,780 |
| $90.00 | 10 | $23,680 |
Five traders with long positions:
| Trader | Position Size | Leverage | Entry Price | Margin | Liquidation Price |
|---|---|---|---|---|---|
| Alice | 100 BTC | 50x | $100 | $200 | $98.00 |
| Bob | 80 BTC | 25x | $100 | $320 | $96.00 |
| Carol | 60 BTC | 20x | $100 | $300 | $95.00 |
| Dave | 50 BTC | 10x | $100 | $500 | $90.00 |
| Eve | 40 BTC | 5x | $100 | $800 | $80.00 |
Total long open interest: 330 BTC ($33,000 notional at $100)
The Trigger: External Sell Pressure
A whale sells 50 BTC at market. This consumes the entire $100 level on the bid side.
Price after trigger: $99.00 (best bid)
No liquidations yet — the closest liquidation price is Alice at $98.00.
Step 1: Alice Gets Liquidated
Price continues to drift down. A few more spot sells push the price to $98.00. Alice's MF drops below MMF.
Alice's liquidation: Exchange must sell 100 BTC.
Liquidation order fills against the order book:
| Level Hit | Size Consumed | Fill Price |
|---|---|---|
| $98.00 | 30 BTC | $98.00 |
| $97.00 | 25 BTC | $97.00 |
| $96.00 | 20 BTC | $96.00 |
| $95.00 | 20 BTC | $95.00 |
| $94.00 | 5 BTC | $94.00 |
Total sold: 100 BTC Average fill: $96.55 New best bid: $94.00 (with 10 BTC remaining at that level) Price impact from Alice's liquidation alone: $98.00 → $94.00 = $4.00 drop (4.1%)
Step 2: Bob and Carol Get Liquidated
The price is now $94.00. This is below both Bob's liquidation price ($96.00) and Carol's liquidation price ($95.00). Both are now in liquidation.
Bob's liquidation: 80 BTC to sell Carol's liquidation: 60 BTC to sell Combined: 140 BTC to sell
Remaining order book:
| Level Hit | Size Consumed | Fill Price |
|---|---|---|
| $94.00 | 10 BTC | $94.00 |
| $93.00 | 15 BTC | $93.00 |
| $92.00 | 10 BTC | $92.00 |
| $91.00 | 10 BTC | $91.00 |
| $90.00 | 10 BTC | $90.00 |
Only 55 BTC of liquidity available to fill 140 BTC of liquidation orders.
The order book is exhausted. 55 BTC fills at an average of ~$92.00. The remaining 85 BTC has no resting bids to fill against.
New best bid: Nothing — the visible book is empty above $90.00 and the $90.00 level is consumed.
Step 3: Dave Gets Liquidated (And the Book Is Gone)
Dave's liquidation price was $90.00. The price has blown through it. Dave's 50 BTC needs to be sold, but there are no resting bids on the visible book.
At this point, the exchange must rely on fallback mechanisms:
- Backstop liquidity providers step in to absorb at a distressed price
- ADL forces profitable short traders to close their positions
- Or the price gaps to wherever the next bid exists
Summary Table: The Cascade
| Step | Event | Sell Pressure | Price Before | Price After | Drop |
|---|---|---|---|---|---|
| 0 | Whale sells 50 BTC | 50 BTC | $100.00 | $99.00 | 1.0% |
| 1 | Alice liquidated | 100 BTC | $98.00 | $94.00 | 4.1% |
| 2 | Bob + Carol liquidated | 140 BTC | $94.00 | < $90.00 | > 4.3% |
| 3 | Dave liquidated | 50 BTC | < $90.00 | ??? | ??? |
| Total | 340 BTC | $100.00 | < $90.00 | > 10% |
The initial trigger was 50 BTC of selling. It resulted in 290 BTC of additional forced selling (Alice + Bob + Carol + Dave). The cascade amplified the selling pressure by 5.8x.
Eve (5x leverage, liquidation at $80.00) survives — but only because she used conservative leverage. If the cascade continued deeper, even she could be at risk.
What This Looks Like Over Time
Price
$100 ┤━━━━━━━┓
│ ┃ ← Whale sells (trigger)
$98 │ ┗━━┓
│ ┃ ← Alice liquidated
$96 │ ┃
│ ┃ (100 BTC eating through bids)
$94 │ ┗━━┓
│ ┃ ← Bob + Carol liquidated
$92 │ ┃
│ ┃ (140 BTC, book nearly empty)
$90 │ ┗━━┓
│ ┃ ← Dave liquidated, no bids
$88 │ ┃ (backstop/ADL territory)
│ ┃
$86 │ ┗━━━━━━━━ price stabilizes
│ (leverage flushed)
└────────────────────────────────── Time
t=0 t=5s t=15s t=30s t=60sNotice the accelerating pattern: each drop is faster and steeper than the last because (a) more positions are clustered nearby, and (b) the book gets thinner as market makers pull quotes.
6. Exchange-Level Defenses
Exchanges have developed several mechanisms to prevent or mitigate liquidation cascades. No single mechanism is sufficient — effective protection requires layering multiple approaches.
6.1 Tiered / Partial Liquidation
Instead of liquidating an entire position at once, the exchange closes it in pieces.
How it works: When an account's MF drops below MMF, the exchange liquidates only a portion of the position (e.g., 10-25%) per cycle. After each partial liquidation, it re-checks the margin. If the partial close was enough to restore MF above the exit threshold, the remaining position is left open.
Why it helps: Reduces the size of individual liquidation orders hitting the book. A 10 BTC liquidation order has far less price impact than a 100 BTC order. It gives the order book time to refill between liquidation chunks.
6.2 Liquidation Throttling
Adding intentional delays or probability gates to the liquidation process.
How it works: Instead of executing liquidations as fast as possible, the engine introduces friction — either a time delay between liquidation batches, or a probability check that randomly skips some execution cycles.
Why it helps: Spreads liquidation pressure over time. If liquidations take 20 seconds instead of 2 seconds, market makers have time to refill the book between waves. Other participants have time to step in as buyers.
6.3 Backstop Liquidity Providers (BLPs)
Designated market makers who contractually agree to absorb liquidation flow when the order book cannot.
How it works: When on-book liquidation fails (not enough resting orders, or price impact would be too severe), the exchange routes the liquidation to registered backstop LPs. These firms take the other side of the trade at a defined price, typically derived from the mark price with a spread.
Why it helps: Provides a guaranteed floor of liquidity specifically for liquidations. Prevents the scenario where liquidation orders eat through the entire visible book and gap the price.
6.4 Auto-Deleveraging (ADL)
Forcing profitable traders on the opposite side to close their positions against the liquidating account.
How it works: As a last resort, the exchange matches the liquidating position against traders who hold opposing positions (and are therefore profiting from the price move). These traders are forced to realize their profit and close their position.
Why it helps: Guarantees that every liquidation can be completed, even when there is zero resting liquidity and backstop capacity is exhausted. Prevents socialized losses and bad debt.
The trade-off: ADL is unpopular because profitable traders lose their position involuntarily. They keep their profit, but lose the opportunity to ride the trade further.
6.5 Insurance Funds
A pool of capital funded by liquidation surpluses that covers shortfalls when liquidating accounts result in bad debt.
How it works: When a liquidation fills at a better price than the bankruptcy price (the price at which the account's equity hits zero), the surplus goes into the insurance fund. When a liquidation results in a loss (bad debt), the insurance fund covers it.
Why it helps: Absorbs losses without requiring ADL or socialized loss. A well-funded insurance fund means ADL is rarely needed.
The risk: Insurance funds can be depleted rapidly during a severe cascade. If the fund is exhausted, the exchange must fall back to ADL or socialized loss.
6.6 Circuit Breakers / Price Bands
Limiting how far the price can move within a given time window.
How it works: The exchange rejects or queues orders that would move the price more than X% from the mark price or index price. Some implementations pause trading entirely for a cooldown period.
Why it helps: Prevents liquidation orders from gapping the price through the entire book. Gives market makers time to re-quote. Prevents manipulation where someone intentionally crashes the price to trigger cascading liquidations.
6.7 Funding Rate Adjustments
Using the funding rate mechanism to discourage crowded positioning before a cascade can form.
How it works: When open interest is heavily skewed (e.g., far more longs than shorts), the funding rate increases, making it expensive to hold the crowded side. This naturally reduces leverage in the system before a shock occurs.
Why it helps: Preventive rather than reactive. By making extreme leverage expensive, the system self-corrects before cascades form. Lower pre-shock leverage means a given price move triggers fewer liquidations.
Comparison of Defense Mechanisms
| Mechanism | Type | When It Acts | Effectiveness Against Cascades | Trade-off |
|---|---|---|---|---|
| Tiered liquidation | Mitigation | During liquidation | High — reduces per-order impact | Slower liquidation, more risk to exchange |
| Throttling | Mitigation | During liquidation | High — spreads pressure over time | Liquidating account remains at risk longer |
| Backstop LPs | Absorption | After on-book fails | High — guaranteed liquidity | Requires LP agreements and capital |
| ADL | Last resort | After backstop fails | Complete — always closes positions | Unpopular, disrupts profitable traders |
| Insurance fund | Absorption | After liquidation | Moderate — covers bad debt only | Fund can be depleted |
| Circuit breakers | Prevention | During price moves | High — stops cascade momentum | Can trap traders, delays price discovery |
| Funding rates | Prevention | Before events | Moderate — reduces buildup | Slow-acting, doesn't help during crash |
7. How Backpack Handles Liquidation Risk
Backpack Exchange implements a multi-layered defense system designed to prevent cascades while ensuring all positions can be safely closed. The system uses a 3-tier liquidation architecture with built-in throttling at every level.
The 3-Tier System
MF Level:
─────────────────────────────────────────
100% Normal trading
│
MMF TIER 1: On-book liquidation starts
│ (IOC orders, throttled)
│
mf_auto_close = max(MMF/2, MMF - 6%)
│
TIER 2: Backstop liquidation
│ (designated LPs absorb)
│
TIER 3: ADL
│ (auto-deleverage vs profitable traders)
│
0% Account equity = 0For full details on each tier, see Trading System Overview and Collateral.
Tier 1: On-Book Liquidation with Throttling
When an account's MF drops below MMF but remains above mf_auto_close, the system liquidates via standard order book matching. The critical design choice is how these liquidation orders are paced:
Throttling mechanics:
- The liquidation loop runs every 1 second
- Each cycle has a 50% probability of executing (a coin flip)
- When it does execute, it liquidates a maximum of 10% of the position
- Expected time to fully liquidate: ~20 seconds
Why this matters for cascades: Compare to an exchange that liquidates 100% of a position immediately. A 100 BTC position on Backpack generates a ~10 BTC sell order every ~2 seconds (on average). On a non-throttled exchange, the same position generates a single 100 BTC sell order. The Backpack approach gives the order book time to refill between chunks, and gives the price time to stabilize.
Liquidation orders are IOC (Immediate-Or-Cancel) limit orders that cross the spread, with a price bound of 2% from the index price. This prevents liquidation orders from filling at absurd prices during a dislocated market.
Tier 2: Backstop Liquidity Providers
If an account's MF drops below mf_auto_close (defined as max(MMF/2, MMF - 6%)), on-book liquidation has failed to restore health fast enough. The system escalates to backstop liquidation.
How Backpack's backstop works:
- Designated market makers register as backstop liquidity providers (BLPs) for specific markets
- Each BLP has a capacity limit that refreshes hourly and minutely
- When triggered, BLPs are contractually obligated to take the other side of the liquidation
- The fill price is derived from the mark price: BLPs receive 2/3 of the spread between the mark price and the account's zero-equity price, with 1/3 going to the liquidity fund
This is a critical layer that most exchanges do not have. On many platforms, the system goes directly from on-book liquidation to ADL. Backpack's backstop layer provides a buffer of guaranteed liquidity that can absorb a significant volume of liquidations without affecting the public order book at all.
Tier 3: Auto-Deleveraging (ADL)
If backstop capacity is exhausted, the system triggers ADL as a last resort.
Backpack's ADL design:
- Selects counterparties by margin fraction ascending (lowest MF traders are selected first)
- Uses a two-phase approach: first tries delta-reducing fills only (reduces the counterparty's position without flipping it), then removes that constraint if insufficient
- Traders using lower leverage (higher MF) are selected last
This means conservative traders are protected. A trader with 2x leverage is far less likely to be ADL'd than a trader with 20x leverage on the same side — the 20x trader has a lower MF and gets selected first.
Tiered Exit Requirements
To prevent liquidation oscillation (account exits liquidation, immediately re-enters, exits, re-enters), Backpack requires accounts to have more margin to exit liquidation than to enter it:
| Account Net Equity | Exit Requirement |
|---|---|
| < $10k | MMF x 1.01 (1% buffer) |
| $10k - $250k | MMF x 1.0075 (0.75% buffer) |
| $250k - $1M | MMF x 1.005 (0.5% buffer) |
| > $1M | MMF x 1.0025 (0.25% buffer) |
Larger accounts get a smaller buffer because their positions are more expensive to liquidate and more likely to cause market impact. The buffer ensures that once a position is partially liquidated enough to restore health, there is a meaningful cushion before it could re-enter liquidation.
How These Layers Work Together Against a Cascade
Consider the walkthrough from Section 5, but on Backpack:
- Whale sells 50 BTC, price drops to $98. Alice's MF drops below MMF.
- Alice enters Tier 1. Instead of dumping 100 BTC at once, the system sells ~10 BTC per cycle, every ~2 seconds. After 5 cycles (~10 seconds), 50 BTC has been sold. The price may have dropped to $96, but the book has had time to partially refill between chunks.
- Bob's MF drops below MMF at $96. Bob enters Tier 1 as well. Now both Alice and Bob are being throttled — the system is processing ~10 BTC from each per cycle, not 100 + 80 BTC simultaneously.
- If the book thins out severely and Alice's MF drops below
mf_auto_close, she escalates to Tier 2. Backstop LPs absorb her remaining position off-book, without touching the public order book at all. This prevents her liquidation from pushing the price down further. - If backstop capacity is exhausted, the system moves to Tier 3 (ADL), matching remaining positions against profitable shorts. This also happens off-book.
The net effect: the cascade from Section 5 that would have crashed the price from $100 to below $90 in seconds on a non-throttled exchange is spread over 20+ seconds on Backpack, with a significant portion absorbed off-book by backstop LPs. The price might still drop — but the feedback loop is dampened, giving the market time to find a real clearing price rather than an artificially depressed one.
8. Systemic Risk: How Cascades Spread Across Exchanges
Liquidation cascades are not contained within a single exchange. The interconnected nature of crypto markets means that a cascade on one venue can trigger cascades on others.
The Cross-Exchange Transmission Mechanism
Exchange A Exchange B Exchange C
┌──────────┐ ┌──────────┐ ┌──────────┐
│Cascade │ │ │ │ │
│starts │ │ │ │ │
│Price: $95│──┐ │Price:$100│ │Price:$100│
└──────────┘ │ └──────────┘ └──────────┘
│ │ │
▼ │ │
┌───────────┐ │ │
│ Index │ │ │
│ Price │──────────────┤ │
│ Updates │ │ │
└───────────┘ ▼ │
┌──────────┐ │
┌──────────────│ Mark │ │
│ │ price │ │
│ │ drops │ │
│ └──────────┘ │
│ │ │
│ ▼ │
│ ┌──────────┐ │
│ │Cascade │ │
│ │starts │──────────────────────▶│
│ │Price: $93│ │
│ └──────────┘ ▼
│ ┌──────────┐
│ │Cascade │
└─────────────────────────────────────────▶│starts │
│Price: $91│
└──────────┘Transmission Channels
1. Index price feeds: Most exchanges calculate their mark price using index prices aggregated from multiple venues. When Exchange A's price crashes due to a cascade, it drags down the index price used by Exchange B, which causes liquidations on Exchange B even though nothing happened on Exchange B's own order book.
2. Cross-exchange arbitrage: Arbitrage bots continuously align prices across venues. When Exchange A's price drops to $95 while Exchange B is at $100, bots will sell on Exchange B (pushing its price down) and buy on Exchange A (which may not help if the book is exhausted). The selling on Exchange B can trigger liquidations there.
3. Shared participants: Large trading firms operate on multiple exchanges. A cascade on Exchange A may impair their capital, forcing them to reduce positions on Exchange B and C as well (either manually or via automated risk management).
4. Collateral contagion: If a token used as collateral across multiple exchanges crashes (as happened with FTT during the FTX collapse), accounts on every exchange holding that token as collateral can be simultaneously impaired.
Mitigating Systemic Risk
Exchanges can reduce their contribution to systemic cascades through several design choices:
Mark price damping: Using a moving average or time-weighted index price rather than a real-time spot price. This means a flash crash on one exchange takes minutes, not seconds, to affect mark prices elsewhere. Backpack uses a 60-second moving average of the spread between mid-price and index price in its mark price calculation.
Index price outlier rejection: Capping individual exchange prices that deviate more than a threshold from the median. This prevents a single exchange's cascade from dominating the index. Backpack fetches prices from 15+ external exchanges and caps outliers beyond 1% from the median.
Liquidation throttling: As described above, spreading liquidation pressure over time reduces the severity of each individual exchange's contribution to the cross-venue cascade.
Backstop absorption: Off-book liquidation mechanisms (backstop LPs, insurance funds) absorb liquidation flow without contributing selling pressure to the public order book, which means they don't directly depress the exchange's own price, which means they don't contribute to cross-exchange contagion.
The Unsolved Problem
Despite all these mechanisms, the fundamental vulnerability remains: crypto is a globally connected, highly leveraged, 24/7 market with no central circuit breaker. There is no equivalent of the NYSE's market-wide halt. Each exchange independently decides how to handle extreme conditions, and their decisions affect each other through shared price feeds.
The best any individual exchange can do is ensure its own liquidation system is as non-disruptive as possible — which is exactly what tiered liquidation, throttling, and backstop layers are designed to achieve.
Key Takeaways
Liquidation cascades are the single biggest structural risk in leveraged crypto markets. They have caused more value destruction than any hack or exploit.
The mechanics are straightforward but the dynamics are nonlinear. Small triggers can produce outsized outcomes because each liquidation makes the next one more likely.
Exchange design matters enormously. The difference between a system that dumps 100% of a position instantly and one that throttles liquidations over 20 seconds can be the difference between a minor correction and a market-wide crisis.
Defense must be layered. No single mechanism is sufficient. Effective protection requires throttling + backstop liquidity + ADL + price bounds, working together.
Conservative leverage is the best individual defense. In the numerical walkthrough, Eve (5x leverage) survived while Alice (50x) was the first casualty. Lower leverage means wider distance to liquidation, which means more room to survive a cascade.
Cascades are a systemic, not individual, problem. Your exchange's liquidation engine affects prices on every other exchange. Responsible liquidation design is not just about protecting your own users — it is about not causing harm to the broader market.