Netting & Collateral: Crushing the Exposure

So far you’ve learned to measure counterparty exposure: it’s the replacement cost if your counterparty defaults right now — $\max(\text{value}, 0)$ — and because the future is uncertain you track its average (EE/EPE) and a bad-case quantile (PFE) through time. You’ve seen the shapes: the swap that humps and amortises to zero, the FX forward that climbs to a maturity peak.

This lesson is about the other half of the job: making that exposure smaller. Banks do not sit on raw exposure and pray. They wrap it in two legal contraptions — close-out netting and a collateral agreement — that, between them, can take a book worth tens of millions of gross exposure and crush it down to a thin residual band measured in days, not years. By the end you’ll know exactly what survives that crushing, because that residual is the only thing the next lessons (CVA and the whole XVA family) get to charge you for.

Close-out netting: one number, not many

Analogy. Imagine you and a friend run a tab against each other all year: you spot them lunch, they cover your cab, you lend them a tenner, they buy the cinema tickets. At year’s end you could demand every single IOU be paid in full both directions — or you could just say “after everything, who owes whom, and how much?” and settle one number. Now imagine the friend goes bankrupt. If every IOU stands alone, the bankruptcy administrator happily collects the ones you owe in full and pays you a few cents on the dollar for the ones they owe. If instead a contract says “on default, tear everything up and settle one net figure,” the administrator can’t cherry-pick. That contract is close-out netting, and the friend is your counterparty bank.

The definition. Under an ISDA Master Agreement, all the trades between two parties that fall inside a given netting set are governed by one contract. On a default event, every trade in the set is terminated (“closed out”) simultaneously, marked to its replacement value, and collapsed into a single net amount owed one way. So your exposure to the netting set is:

$E_{\text{net}} = \max\!\left(\sum_i V_i,\ 0\right)$

— the max-of-the-sum — and not the sum-of-the-maxes, $\sum_i \max(V_i, 0)$, which is what you’d be stuck with if each trade defaulted on its own.

Worked example. A netting set of three trades, valued from your side:

Without enforceable netting, each trade stands alone. The two positive trades (A and C) are claims you have on the defaulter — worth +$5m and +$2m, a gross $7m owed to you. But trade B is −$3m: you owe them $3m, and on their default the administrator will absolutely make you pay it in full. Meanwhile your $7m claim goes into the bankruptcy queue and pays back only a fraction. That asymmetry — pay your debts in full, collect theirs at a discount — is the cherry-picking that netting kills. Your exposure is the gross positive:

$E_{\text{no net}} = \max(5,0) + \max(-3,0) + \max(2,0) = 5 + 0 + 2 = 7$

— that is, $7m owed to you.

With close-out netting, the −$3m you owe is allowed to offset the $7m owed to you before anyone defaults out the door:

$E_{\text{net}} = \max(5 - 3 + 2,\ 0) = \max(4, 0) = 4$

— that is, $4m.

The exposure dropped from $7m to $4m. The netting benefit is the slice you saved:

$\text{Netting benefit} = 1 - \frac{E_{\text{net}}}{E_{\text{no net}}} = 1 - \frac{4}{7} \approx 0.43,$

a 43% cut to the number at risk — for the price of a signature, no extra capital, no collateral yet. We define the netting factor as the ratio that survives:

$\text{Netting factor} = \frac{E_{\text{net}}}{E_{\text{no net}}} = \frac{4}{7} \approx 0.57.$

The netting factor lives in $[0, 1]$: it’s 1 when every trade points the same way (nothing to offset — netting does nothing) and approaches 0 when a book is beautifully balanced (offsetting positions cancel almost perfectly). Real dealer books, stuffed with offsetting hedges, often net down by 70–90%.

When it matters

Netting is the first and cheapest exposure reducer, and it’s the foundation everything else sits on: collateral, regulatory capital, and CVA are all computed at the netting-set level, on the net number. A trade’s marginal contribution to exposure can even be negative — adding a hedge that offsets the book reduces total exposure, which is why the same new trade can carry a very different CVA charge depending on which netting set it lands in. Get the netting set wrong (or lose the legal opinion) and every downstream number inflates.

The CSA: collateral that moves with the mark

Analogy. Netting shrinks the exposure; collateral chases what’s left to roughly zero, every single day. Think of a security deposit on a flat that gets re-assessed nightly. Normally a deposit is posted once and sits there. A CSA deposit is alive: every evening someone re-values the relationship, and whoever is now “ahead” tops up — or claws back — the deposit so that neither side is ever owed much. The mark drifts in your favour today? They wire you cash tonight. It drifts back tomorrow? You return some. The pile of collateral shadows the mark like a margin account on a brokerage, because that’s exactly what it is.

The definition. A Credit Support Annex (CSA) is the collateral agreement bolted onto the ISDA Master. (Master = the netting rulebook; CSA = the collateral rulebook attached to it.) It specifies who posts what, how often, in what currencies, at what valuations, and with what frictions. Its job: as the net mark-to-market of the netting set moves in one party’s favour, the other party posts collateral — overwhelmingly cash, sometimes high-grade bonds — so that the collateralised exposure stays near zero instead of drifting up into a hump or a ramp.

So the running order is always: net the book first (close-out netting gives one number), then collateralise that net number (the CSA tops it to ~zero). Collateral doesn’t replace netting; it stands on its shoulders.

Variation margin (VM): marking exposure to ~zero

Analogy. VM is the nightly settle-up on a sports bet between friends who don’t trust each other to pay at the end. Instead of letting a season’s worth of wins pile into one giant IOU, they settle every night: lost ground today, you hand over tonight’s difference in cash. Nobody is ever owed more than a single day’s swing. That’s variation margin — the running score, paid in cash, every day.

The definition. Variation margin (VM) is collateral posted to cover the current net mark-to-market of the netting set, exchanged on a regular cycle (daily for most uncleared portfolios, intraday for cleared ones). When the net mark moves $1m in your favour, the counterparty posts $1m of VM; when it moves back, the VM is returned. Because the collateral tracks the mark step for step, the running collateralised exposure collapses from the full hump-or-ramp profile down to a thin band. VM is two-way and fungible: it can be re-used (rehypothecated), and it’s title-transferred, so the poster generally loses legal title to it.

This is precisely what the collateral checkbox on the exposure island below switches on. Tick “Add collateral (CSA)” and watch what happens to both profiles: the interest-rate-swap hump and the FX-forward ramp — those big multi-year arcs of uncollateralised exposure — both collapse to a thin, near-flat band hugging the axis. The band isn’t quite zero (we’ll see why in the next section), but the towering exposure is gone. Toggle it on and off a few times on both instruments to feel how violently collateral flattens the picture.

Initial margin (IM) and the margin period of risk

Analogy. VM is settling the score every night; but a defaulter doesn’t politely settle on their last night. They miss a call, then there’s a grace period, then notices, then a few days of legally closing out and re-hedging — and over that whole stretch the mark keeps wandering, uncovered. It’s the difference between “we settle nightly” and “what happens during the messy week after you stop answering my calls.” Initial margin (IM) is the airbag you inflate before the crash for exactly that messy week.

The margin period of risk (MPoR). Even flawless daily VM leaves a gap. Between the counterparty’s last met margin call and the moment you’ve fully closed out and re-hedged, time passes: you have to notice the miss, serve notice, wait out cure periods, then liquidate and replace the positions. That window is the margin period of risk — conventionally taken as about 10 business days for uncleared trades (and shorter, ~5 days, for cleared). Over those ~10 days the mark can move against you with nothing fresh posted to cover it. The MPoR is the leftover band you saw collapse-but-not-to-zero on the island.

The definition. Initial margin (IM) is an extra buffer posted up front, independent of the current mark, sized to cover that MPoR gap. Formally it’s a high quantile (typically 99%) of the potential change in the netting-set value over the MPoR:

$\text{IM} \approx \text{quantile}_{99\%}\big(\,\Delta V \text{ over the MPoR}\,\big).$

So while VM answers “what’s the mark right now?”, IM answers “how bad could the move get during the ~10 days it takes to close them out?” Post both, and even a default mid-week is largely covered.

Worked example. A netting set is fully VM’d, so today’s mark is collateralised to ~zero. You estimate the net value has a daily volatility of $1m. Over a 10-day MPoR, the standard deviation of the move scales with $\sqrt{t}$:

$\sigma_{\text{MPoR}} = 1 \times \sqrt{10} \approx 3.16$

— so about $3.16m of standard deviation over the window.

A 99% one-sided quantile of a normal move is about 2.33 standard deviations, so:

$\text{IM} \approx 2.33 \times 3.16 \approx 7.4$

— roughly $7.4m of initial margin.

You’d hold roughly $7.4m of initial margin so that, in 99% of close-out scenarios, the move over the messy 10-day window is already covered before you start. Note the $\sqrt{t}$: doubling the MPoR to 20 days only raises IM by $\sqrt{2} \approx 1.4\times$, not $2\times$.

The standard model — ISDA SIMM. Nobody re-derives that quantile per portfolio from scratch; the industry uses the ISDA SIMM (Standard Initial Margin Model), a common, sensitivity-based recipe so both sides compute the same IM number and don’t argue. Regulation (the uncleared margin rules, UMR) phased in mandatory two-way IM for the largest dealers.

When it matters

IM matters most where VM is weakest: long-dated, volatile, or one-directional books where the 10-day move can be large, and for exactly the systemic dealers UMR targets. For a tightly hedged, low-vol book, IM is small; for a concentrated directional bet, IM can dwarf the VM. And because IM is segregated and funded separately, it carries its own cost — which, foreshadowing hard, is the entire reason MVA (margin valuation adjustment) exists later in this course.

Thresholds, MTA and haircuts

Analogy. A CSA is not a perfectly frictionless cash hose. It has three knobs that let a little exposure leak through for the sake of convenience and trust — like a phone bill with a $50 fee-free allowance (the threshold), a “we won’t bill you for under $5” rule (the MTA), and a “we’ll only credit a coupon at face value if it’s gold-plated” discount on what you pay with (the haircut). Each one is a small, deliberate hole in the otherwise-airtight collateral.

The definitions.

• Threshold. An amount of uncollateralised exposure the parties agree to tolerate before any collateral changes hands. It’s effectively an unsecured credit line you extend to the counterparty — exposure can drift up to the threshold for free, and only the excess gets collateralised. A strong counterparty might get a $10m threshold; a shaky one gets zero (a “zero-threshold CSA,” fully collateralised from the first dollar).
• Minimum transfer amount (MTA). An operational floor: don’t bother wiring collateral for trivial changes. If the required move is below the MTA, nobody transfers anything — it just saves both back offices from settling $4,000 every afternoon. Once the cumulative shortfall crosses the MTA, the whole amount is called.
• Haircut. When collateral is posted as securities rather than cash, the receiver credits it at less than market value to cushion against the collateral’s own price moving before it can be sold. Post $100 of a government bond at a 2% haircut and you’re credited only $98; you must over-post to cover a given exposure. Riskier or more volatile collateral gets a bigger haircut (equities far more than T-bills).

Worked mini-example. A CSA has a $5m threshold and a $1m MTA. Your net exposure to the counterparty climbs to $5.4m.

• The first $5m sits inside the threshold — uncollateralised by agreement. So the collateral required is only the excess: $5.4m − $5m = $0.4m.
• But $0.4m is below the $1m MTA, so no collateral moves at all — the call is too small to bother with.
• Net result: you’re carrying the full $5.4m uncollateralised right now, even with a CSA in force. Push exposure to $6.2m and the excess is $1.2m, which clears the MTA, so the counterparty now posts the whole $1.2m (not just the bit over $1m).

Now suppose they post that $1.2m as a bond at a 2% haircut: $1.2m of bonds credits only $1.2 \times 0.98 = 1.176$ (i.e. $1.176m), so they’d actually have to post about $1.2 / 0.98 \approx 1.224$ (i.e. $1.224m) of bonds to cover the $1.2m call. The threshold, the MTA and the haircut each leave a little exposure on the table — by design.

When it matters

Thresholds and MTAs are exactly why a “collateralised” trade still carries residual CVA: the threshold is uncollateralised by contract, and the MTA means small shortfalls go uncovered. The weaker the counterparty, the more a desk drives the threshold and MTA toward zero — the strongest protection is a zero-threshold, low-MTA, cash-only CSA. Haircuts matter most when bond or equity collateral is allowed, where a stressed market can move the collateral and the exposure together.

What’s left after netting + collateral

You’ve now applied the full crush, in order. Start with gross exposure across the book; net it to one number per netting set (a 40–90% cut for a signature); then collateralise that net number with daily VM (years of exposure squeezed to days) plus segregated IM (the airbag for the close-out window). What can possibly be left?

A thin, stubborn residual, and it’s the sum of the deliberate leaks:

$E_{\text{residual}} \;\approx\; \text{MPoR gap} \;+\; \text{threshold} \;+\; \text{haircut slippage}$

— the mark drifting over the ~10-day MPoR gap, plus the threshold that’s uncollateralised by contract, plus the haircut slippage on any non-cash collateral.

That’s it. That’s the band hugging the axis on the island. The towering multi-year hump is gone; what remains is a few days of margin-period drift, plus whatever exposure the threshold and MTA waved through, plus the slack a haircut left on the collateral. For a zero-threshold, cash-only, daily-VM, IM’d netting set, the residual is almost entirely the MPoR gap — which is why “what’s the margin period of risk?” is the question that sets the size of every collateralised CVA.

And that’s the punchline of the whole lesson: this residual is exactly what gets priced next. CVA — and the entire XVA family (DVA, FVA, MVA, KVA) that follows it — is integrated against this number, not the raw gross exposure. Netting and collateral don’t just reduce your risk; they define the surface that every valuation adjustment is computed on.

Next — CVA: the price of default. You’ve crushed the exposure down to a thin residual; now we put a price on it. CVA takes that residual exposure profile, multiplies it by the counterparty’s probability of defaulting and by how little you’d recover if they did, and discounts the lot back to a single number — the market value of the credit risk you’re carrying. The residual you built here is the raw material; CVA is the bill.