Kinetic Alpha

Research · Derivatives · Margin & liquidation

Offshore perpetual futures — margin, liquidation, and the October 10 stress test

On October 10-11, 2025, $19 billion in leveraged crypto positions were liquidated in hours — the largest single cascade in market history. 1.62 million accounts. 87% of the positions were longs. The Trump tariff announcement triggered an S&P 500 −2% move inside ten minutes; BTC fell from $122k to $105k. Binance tapped roughly $188M of its insurance fund. Hyperliquid force-closed $10B+ in positions — its first ADL event in two years. Coinbase Derivatives was untouched because its regulated 10× leverage cap kept the cascade from reaching it.

That 36-hour window was the real-time stress test of every major offshore perpetual futures venue's design choices: margin methodology, liquidation engine, mark price sourcing, insurance fund sizing, ADL ranking logic, transparency. Each design performed differently. This piece works through what each venue actually does at the rulebook level and what that means for both the user holding a position and the exchange running the book.

The interactive comparison dashboard below has four views: a side-by-side matrix across 18 dimensions, an Oct 10 stress-test breakdown, a flash-crash simulator, and a risk scoring view.

Drive the comparison

Interactive comparison dashboard

Toggle any subset of the seven venues, compare across 18 dimensions, simulate a flash crash at chosen leverage and price drop, and see the qualitative risk scoring.

Loading comparison dashboard…

Section 1

The competitive landscape — June 2026

The offshore perp landscape is two-tier and increasingly fragmented. On the centralized side, Binance remains dominant — roughly 29-30% of Bitcoin futures open interest, the single largest venue by every metric. Bybit (~13-14%) and OKX (top three) anchor the second tier. Gate.io and Bitget are competitive in the third tier with comparable mechanics.

On the decentralized side, the picture shifted dramatically in 2025-2026. Hyperliquid cleared $619 billion of perp volume in Q1 2026 alone — the dominant on-chain perp DEX by a wide margin. Aster ($319B) and Lighter ($254B) are credible challengers; combined perp DEX volume hit $2.41 trillion in Q1 2026, capturing roughly 26% of the global derivatives market by some estimates. dYdX (now on its own Cosmos chain after v4) remains structurally important but no longer the volume leader.

The regulated comparison set is small but growing. Coinbase Derivatives launched CFTC-regulated perpetual-style futures in the US on July 21, 2025 — nano BTC and ETH contracts, 10× leverage, 24/7, CFTC-DCM rulebook. Coinbase also closed its $2.9B acquisition of Deribit in August 2025, becoming the global leader in crypto derivatives by open interest and options volume. US clients gained no-action-letter access to Deribit options (and shortly perps) via Coinbase Bermuda starting May 2026.

Section 2

Margin methodologies — how leverage is rationed

Every venue rations leverage with some combination of initial margin (IM), maintenance margin (MM), and tiered scaling. The structural differences across the seven venues cluster into four patterns.

Pattern 1: Tiered IM/MM with cross / isolated / portfolio modes (Binance, Bybit, OKX)

The CEX standard. A trader picks Cross (single margin pool across positions), Isolated (margin walled per position), or Portfolio Margin (risk-based netting across the full account). Within each mode, leverage is gated by tiered MM — Binance's BTC ladder runs from 125× at the smallest notional tier down to 1.25× at the largest, with MM rising from 0.40% to 5%+. Bybit and OKX use comparable ladders with slightly less aggressive top-end leverage (100× and 100× respectively).

The key implementation differences sit in portfolio margin design. Bybit's UTA (Unified Trading Account) and OKX's multi-currency portfolio mode use stress-test-based margin calculation — a scenario shock applied to all positions simultaneously, with MM set at the worst-case PnL. This is the most capital-efficient mode but requires the largest stress-window calibration. Binance's portfolio margin is similar in concept but applied across its USD-M, Coin-M, and spot books in a unified hierarchy.

Pattern 2: Per-asset max-leverage caps (Hyperliquid)

Hyperliquid does not use notional-tier scaling. Instead, each listed perpetual has a fixed max leverage (3× to 40× depending on liquidity and volatility), and MM is set at exactly half of IM at that max. This produces an asset-by-asset MM range from 1.25% (40× assets like BTC) to 16.7% (3× assets like illiquid alts). The simplicity is a feature — there's no tier-down to navigate, and the rules are uniform across position sizes. The trade-off is capital efficiency at large size: a 100M-notional BTC position posts the same MM as a 100K position, which on a tiered-CEX venue would post much more.

Pattern 3: IMF scaling with open notional (dYdX v4)

dYdX takes the opposite approach: Initial Margin Fraction scales linearly with the total open notional in a perpetual market — not with the trader's individual position size, but with the market's overall liquidity stress. As open notional approaches a cap, IMF rises until it caps at 100% (1:1 collateral required to add positions). This is a governance-controlled circuit breaker against runaway leverage in individual markets, unique among the venues.

Pattern 4: Regulated SPAN-family (Coinbase Derivatives, Deribit)

Both use scenario-based margin calculation derived from CME's SPAN methodology. Coinbase Derivatives caps leverage at 10× by design — the regulated answer to "what's a safe maximum?" Deribit's portfolio margin is mature on options portfolios, with Greeks netting that the offshore CEX venues haven't quite matched. Both run via clearing-house structures that mutualize loss through guarantee funds rather than insurance funds and ADL — a fundamentally different risk distribution.

Section 3

Liquidation engines — how positions are forced out

Liquidation is the bit that actually matters. When mark price reaches the liquidation price, the engine has to close the position — and the design choices about how, where, and at what cost determine whether the user loses just their margin or also gets caught in slippage cascades, ADL, or socialized loss.

The mark price formula

All five offshore venues use some variant of the same median-based mark price:

Mark = Median (Index × (1 + Funding × t/8), Index + 5-min basis MA, Last Trade Price)

The three-input median is the standard defense against manipulation-driven liquidations — no single input can drive the mark away from the others. The Index input itself is typically composite across 5-10 major spot exchanges with outlier rejection. Binance, Bybit, OKX, and Bitget all use close variants of this formula. Hyperliquid's implementation is on-chain (oracle index + EMA of own book + 1-min impact mid) but the structural logic is the same. dYdX uses Pyth oracle prices directly.

Partial vs full liquidation

All major venues now support partial liquidation — close just enough position to restore MM compliance, not the whole book. The exception in practice is Hyperliquid, which attempts a full-size market order first; partial closure only happens if the book absorbs partial fill. dYdX makes partial liquidation a governance parameter.

Liquidation penalty

The penalty (fee charged on liquidation) varies materially. dYdX caps it at 1.5% by governance default. Binance, Bybit, OKX charge tier-dependent penalties from 0.5% to 5%. Hyperliquid charges no explicit penalty — the execution slippage during forced market orders is the cost. Coinbase Derivatives' FCM-managed close-out uses standard futures-clearing close-out fees, structurally lower than offshore.

Order routing during liquidation

This is where the designs diverge most sharply. CEX venues (Binance, Bybit, OKX) route liquidations to the live order book, absorbing any residual into the insurance fund and then triggering ADL if insurance is exhausted. Hyperliquid routes through a unique two-stage backstop: order book first, then if equity drops below 2/3 of MM, the HLP vault (LP-funded backstop pool) absorbs the position at the mark-based liquidation price. dYdX generates a protocol-side liquidation order matched against the live book at a "Fillable Price" — closer to the CEX pattern but with on-chain settlement. Coinbase / Deribit use FCM/clearing-house managed close-outs with guarantee-fund mutualization as the last resort.

Section 4

Insurance funds and ADL — the loss-absorption layer

When liquidation order routing fails to close a position at or above the bankruptcy price, the residual loss has to land somewhere. The two main mechanisms are insurance funds (exchange-funded, with quarterly contributions from liquidation penalties) and auto-deleveraging (ADL — force-close the most profitable, most leveraged opposing positions to absorb the loss). Both have hard limits, and both came under stress on Oct 10.

Insurance fund sizing

Binance's insurance fund is the largest in the offshore space — typically $1B+ across its USD-M and Coin-M markets. Bybit's fund sits around $300-500M. OKX discloses an aggregated risk reserve quarterly. Hyperliquid's HLP vault (which functions as both a liquidator-of-last-resort and an LP-yield product) holds >$300M of deployed capital and its insurance fund is on-chain transparent. dYdX's insurance fund size is visible on-chain in real time. Deribit's fund is smaller (~$50M+) because portfolio-margin requirements keep individual position risk much lower.

How ADL ranks positions

ADL is the exchange's last resort. The ranking algorithm determines whose positions get force-closed. The standard ranking is profit percentile × leverage percentile — the most profitable AND most leveraged opposing positions go first. Binance, Bybit, and OKX all use close variants of this. Hyperliquid's ranking is on-chain visible. dYdX's invocation is rare enough that it didn't fire in Oct 10. Coinbase Derivatives doesn't use ADL at all — cleared products use guarantee-fund mutualization instead.

What Oct 10 actually showed

Binance tapped ~$188M of insurance and invoked ADL across multiple symbols — the largest absolute insurance draw. Hyperliquid's HLP absorbed billions of dollars of liquidations cleanly but eventually had to invoke its first ADL event in two years. The on-chain transparency meant every action was auditable; the trade-off was that the size of the ADL was visible to everyone in real time, which can amplify panic. Coinbase Derivatives' US-only regulated structure was effectively immune because the 10× leverage cap meant no clean cascade developed.

Section 5

Funding rate mechanics — where the perp anchors to spot

The funding rate is the perpetual-product mechanism that keeps the contract anchored to spot. Three patterns dominate:

  • Premium-only, 8-hour cadence (Binance, Bybit, OKX, Bitget, Gate.io). Funding = clamped premium index, paid every 8 hours (00:00 / 08:00 / 16:00 UTC). Funding caps are typically ±0.75% per period for BTC, higher for alts. The 8-hour cadence creates predictable funding-event windows that traders can game.
  • Premium-only, 1-hour cadence (Hyperliquid, dYdX). Same formula but hourly. Smoother in calm regimes, more capital-efficient in trending regimes. The shorter cadence reduces the size of single funding events but increases frequency.
  • Cash-carry style (Coinbase Derivatives, CME spot-quoted). Funding = SOFR + spread, no premium-only term. The contract anchors to spot via daily financing adjustment rather than a market-driven premium. Slower to respond to extreme positioning but conceptually closer to traditional futures.

The structural implication: offshore venues with premium-only funding can see extreme funding-rate moves (Hyperliquid's per-hour cap is ±4%, which annualizes to over 35,000% in extreme cases). Regulated venues using cash-carry can't price extreme positioning into funding — they rely on margin requirements to constrain it.

Section 6

Risk through the user's lens

Five distinct risk categories show up for the user:

  • Position-side risk. 100× leverage on Binance means a 1% adverse move wipes out your margin (less, after fees). The same trade at 10× on Coinbase Derivatives requires a 10% move. Higher leverage is offered as a feature; in practice it accelerates total margin loss.
  • Slippage during liquidation. Even when the calculated liquidation price is reasonable, the actual fill during a cascade often happens much worse — the difference between bankruptcy price and execution price is what gets pulled from insurance funds. On Oct 10, this is what consumed Binance's $188M draw.
  • ADL risk on the surviving side. If you're profitable, highly leveraged, and on the right side of a cascade, your position may still get force-closed via ADL — capping your win. Hyperliquid's transparency means you can see your ADL rank in advance; CEX venues do not.
  • Counterparty / withdrawal risk. Offshore CEX venues have historically imposed withdrawal halts during stress. Hyperliquid and dYdX don't have this risk because user funds sit in self-custody on-chain. Coinbase Derivatives runs through a regulated FCM with US bankruptcy protections.
  • Regulatory / access risk. US users can't legally access Binance, Bybit, OKX directly. They can access Hyperliquid (no KYC), dYdX (no KYC), and Coinbase Derivatives (full KYC). The Deribit/Coinbase Bermuda no-action structure opens US access to Deribit options and (soon) perps under a CFTC-supervised pathway.

Section 7

Risk through the exchange's lens

The exchange-side risks are different and arguably more consequential at scale:

  • Insurance fund exhaustion. Every CEX's insurance fund is finite. When a cascade overruns it, the exchange has to choose between ADL (force-close opposing positions, eating user PnL) and absorbing the loss to its own balance sheet (which can threaten solvency). The Oct 10 $188M Binance draw was within tolerance; a 5× event would not be.
  • Socialized loss / clawback risk. Some venues historically used clawback (taking PnL from profitable users post-hoc to cover bad debt). This has been largely replaced by ADL but creates user-trust risk if invoked.
  • Mark price manipulation exposure. The three-input median mark price is defense in depth, but isolated thin alts have been manipulated to trigger cascading liquidations (Mango Markets 2022 is the canonical case). Each venue's listing process is its first line of defense.
  • Regulatory enforcement risk. Binance settled with the US DOJ for $4.3B in 2023. OKX faces ongoing inquiries. Bybit has restricted US access. The offshore-CEX model is operating under accelerating regulatory scrutiny. The DEX-on-chain model carries different but real risk: SEC / CFTC enforcement actions against frontends and developers continue. Coinbase Derivatives is the only category with full US regulatory cover.
  • Platform reliability under stress. Oct 10 saw Binance struggle with platform load; stop-losses failed for some users; accounts froze briefly. Hyperliquid ran cleanly because the chain processes orders deterministically. Reliability is a moat the offshore CEX venues are still building.

Section 8

The regulated alternative — where Coinbase fits

Coinbase has built the only regulated US perpetual-futures alternative to the offshore CEX/DEX market. Three pieces fit together:

  • Coinbase Derivatives (DCM) launched nano BTC and ETH perpetual-style futures in July 2025 — CFTC-regulated, 10× max leverage, 24/7 trading, cash-carry financing rather than premium-only funding. Fee structure as low as 0.02%. The regulated counterpart is the perpetual analog of the existing nano CME contracts.
  • Deribit acquisition ($2.9B, closed August 2025) gave Coinbase the global leader in BTC/ETH options open interest. Deribit retains its offshore (Panama + Bermuda) regulatory footprint but as a Coinbase subsidiary now interoperates with the regulated US business.
  • CFM / CBBM access pathway — Coinbase Financial Markets (a CFTC FCM) received no-action relief in May 2026 permitting it to post US customer crypto collateral with Coinbase Bermuda Limited to margin foreign futures and options positions on Deribit. This is the first regulated bridge giving US clients access to the deep Deribit options book. Perp access is on the same pathway.

The structural argument: for US-domiciled users, the Coinbase regulated stack is now roughly comparable in leverage to onshore CME-cleared products (10× regulated cap) but with 24/7 trading, perpetual settlement, and growing collateral flexibility. The offshore alternative offers materially higher leverage, broader asset coverage, and (in the DEX case) self-custody — but with the regulatory access risk and the Oct 10 stress-test reality of insurance fund exhaustion and ADL invocation.

Section 9

What to watch for the next stress test

Three things to watch — they'll determine how the next Oct-10-scale cascade resolves:

  • Insurance fund growth vs leverage growth. If insurance funds are scaling proportionally to OI growth, the system is becoming more resilient. If OI is outrunning insurance, the next ADL event will be larger than Oct 10. Insurance fund size publication discipline is uneven — Binance and Hyperliquid publish, others less.
  • ADL ranking transparency. Hyperliquid's on-chain ADL rank is visible to users in real time, which lets them de-risk before getting force-closed. The CEX venues do not provide this — your rank is opaque until you're hit. This is a structural transparency gap that's tradeable if the venues close it.
  • Cross-venue contagion paths. Oct 10 stayed contained: no exchange failed, no major lender went down. The next event might not. Margin financing across CEX venues (where market-makers borrow from one venue's spot to provide perp liquidity at another) creates cross-venue contagion paths that aren't well documented. The single largest open question for any institutional allocator: who's the counterparty to your counterparty?

Sources and provenance

Companion work

Daniel Kaufman · Kinetic Alpha · June 2026. Research and education only. Exchange parameters (insurance fund sizes, leverage tiers, MM ratios) are illustrative ballparks calibrated to published documentation as of June 2026 and subject to change. Re-mark against live exchange docs before any sizing decision. Not investment advice. Contact: dkaufmanrisk@gmail.com.