Margining and Liquidations, Part 3: Scenario-Based Margining

This is part three of our Margining and Liquidations series. Read Part 1 for isolated markets and Part 2 for cross markets. This part picks up exactly where Part 2's closing subtlety left off: hedges don't net under additive margining, even though your net risk is close to zero. Here we look at how venues actually solve that.

The idea

Same relationship as always: $\mathsf{MM}$ is what you need to survive, $\mathsf{IM}$ is what you need to open, and $\mathsf{MM}\le \mathsf{IM}$. What changes here is how the underlying number gets computed.

Define a shared set of scenarios $s_j$, each assigning a hypothetical P&L to every position ("BTC drops 15%, IV spikes 20 points," say). The joint worst case across those scenarios is:

That's the number a scenario grid actually computes directly: the minimum collateral to survive the worst shock in the set. $\mathsf{IM}$ is this same number plus a buffer, so opening a position doesn't put you right at the liquidation edge:

The buffer can be a flat multiplier, like Deribit's old $\mathsf{IM}=1.2\times \mathsf{MM}$, or its own additive charges, like Deribit's current Delta Shock and Roll Shock (more on this below). A hedge that's genuinely offsetting cancels out scenario-by-scenario, not just on average, so it earns a real credit on $\mathsf{MM}$, and by extension $\mathsf{IM}$, instead of two independent charges.

Revisiting Part 2's hedge

Recall the example: long 1 BTC-PERP and short 1 BTC-PERP-equivalent, both at 10x, BTC at $100. Additive margining charged $\mathsf{MM}=\$20$ ($10 per leg) despite flat P&L. Under a simple scenario set (say BTC ±10%, applied identically to both correlated instruments):

Worst case across both is $0, so $\mathsf{MM}\approx \$0$: a real credit, not just a smaller charge. (In practice it's never quite zero; more on why below).

A crypto-native example: Deribit

Deribit's Portfolio Margin is a good concrete instance of this. Positions are grouped into a risk unit per settlement currency: every BTC-settled instrument (spot, futures, options) shares one unit, ETH another, and so on. $\mathsf{MM}$ comes directly from a price shock scenario grid, roughly ±17.5% for BTC/ETH (wider for smaller-cap currencies), plus an extended tail table out to −66%/+500% to catch deep gap risk.

$\mathsf{IM}$ used to be a fixed multiple of that: $\mathsf{IM}=1.2\times \mathsf{MM}$. Today it's computed independently, with its own additive charges, a Delta Shock and a Roll Shock, layered on top of the core scenario result, exactly the $\mathsf{IM}=\mathsf{MM}+\mathsf{buffer}$ relationship above, just with a more elaborate buffer than a flat multiplier. That layering matters: a scenario grid alone can miss risk that only shows up between shocks or across expiries (a short option that's barely OTM at every grid point but not in between, say), so venues charge a floor on top rather than trusting the grid to catch everything. OKX and Bybit's portfolio margin products follow the same two-part shape, a joint scenario grid plus additive charges for what the grid alone doesn't catch.

Not every "portfolio margin" is this

Deribit, OKX, and Bybit's portfolio margin products are genuine joint scenario computations. Hyperliquid's and Binance's "Portfolio Margin" products are a different thing entirely: a bigger additive pool with cross-collateral and auto-borrowing, but the $\mathsf{MM}$ underneath is still a strict per-position sum: no scenario grid, no netting. Same name, unrelated mechanism, and we'll see both kinds in Part 4's comparison.

Closing

And that’s it! We’ve now covered the three main margining methodologies. We’ve seen how the same parameters (Initial Margin and Maintenance Margin) are computed for these, and how they relate to the account’s equity to determine opening size limits and liquidations.

Part 4 is a wide, detailed survey of how some of the top venues (including top CEXs and DEXs) implement all of this.