Reader question: Two banks agree that a derivatives portfolio has changed value overnight. How does that mark-to-market difference become an actual collateral call, and why can two systems disagree even when both start from the same trades?
A bilateral variation-margin process is a state-reconciliation algorithm. It compares the current value of a legally defined portfolio with the collateral that should already be supporting that exposure, applies the rules in the credit support agreement and relevant regulation, and produces one of three outputs: deliver collateral, return collateral, or do nothing. The difficult part is not subtraction. It is ensuring that both parties are subtracting the same things, at the same valuation time, under the same legal scope and collateral rules.
Page role and safety boundary
This is a computational and mathematical explanation of collateral mechanics for derivatives. It is not legal advice, trading advice or guidance for negotiating an ISDA agreement. Actual transfer amounts depend on the governing master agreement, CSA/CSD terms, applicable margin regime, eligible-collateral schedule, valuation agent provisions and operational procedures.
Variation margin is about current exposure, not future exposure
The first distinction is conceptual. Variation margin (VM) responds to changes in current mark-to-market exposure. Initial margin (IM) is designed to cover potential future exposure during a close-out period. A system that mixes the two without an explicit contractual reason can produce a numerically neat but conceptually wrong call.
For uncleared derivatives, the BCBS–IOSCO margin framework requires covered entities to exchange variation margin under applicable implementation rules. The 2016 ISDA VM Credit Support Annexes were designed specifically to help parties document regulatory variation-margin arrangements. Local rules and contract elections determine the operational detail.
A useful generic state equation
Because CSA wording and sign conventions differ, it is safer to describe the engine generically than to pretend there is one universal spreadsheet formula. Let:
- E = current net exposure under the covered netting set;
- C = recognised collateral already held, after agreed valuation and haircuts;
- P = collateral already agreed but still in transit, if the agreement and operations treat it that way;
- T = any applicable threshold under the agreement or regime;
- MTA = minimum transfer amount;
- R = contractual rounding increment.
A conceptual target-collateral amount is first derived from exposure and the relevant threshold rules. The engine then compares that target with recognised collateral and, where appropriate, pending transfers. The residual becomes a raw delivery amount or raw return amount. Only then are minimum-transfer and rounding rules applied.
In compact form:
raw transfer = target collateral(E, agreement) − recognised collateral(C, P, haircuts).
If the absolute raw transfer is below the contractual MTA, the call may be suppressed. If it exceeds the MTA, the amount can then be rounded according to the agreement. The sign determines who delivers and who receives.
A worked example
Suppose Party A has $12.40 million of net current exposure to Party B under the covered portfolio. Assume, only for this example, a zero VM threshold. A already holds $10.95 million of recognised collateral after the applicable collateral valuation and haircuts. A further $0.30 million has been agreed and is in transit, and the operational rules treat that amount as pending collateral for call calculation.
The uncovered amount is:
$12.40m − $10.95m − $0.30m = $1.15m.
If the agreed minimum transfer amount is $0.50 million, the residual is large enough to generate a call. If the agreement rounds transfer amounts to the nearest $0.10 million, the operational call might become $1.20 million, depending on the exact rounding convention. If the residual had been $0.40 million instead, the MTA could suppress the transfer until later exposure movements push the residual beyond the threshold for action.
This simple example already reveals three sources of disagreement: whether pending collateral counts, whether collateral is valued before or after a haircut, and exactly how rounding is applied.
Why regulatory VM often has a zero threshold
For covered non-centrally cleared derivatives, the global margin framework is designed so current exposure is collateralised rather than left as unsecured credit simply because it is below a negotiated credit threshold. Regulatory VM therefore generally operates with zero threshold for covered transactions, subject to the detailed local rules. The framework also permits a de minimis minimum transfer amount, subject to a cap and local implementation, to prevent operationally inefficient tiny movements.
This distinction matters: threshold changes how much unsecured exposure can remain; MTA changes when an otherwise required movement becomes operationally large enough to transfer. They are not the same parameter.
Collateral is not simply its screen price
If cash in the collateral currency is eligible, valuation can be straightforward. Securities introduce more machinery. The system may need market price, quantity, accrued interest, eligibility rules, concentration limits, currency conversion and a haircut. A $10 million security position with a 2% applicable haircut contributes only $9.8 million of collateral value for the call calculation.
An FX mismatch can introduce an additional haircut under some regulatory schedules. Wrong-way-risk restrictions can make otherwise high-quality collateral unacceptable. The collateral engine therefore has two separate jobs: price the asset and decide how much of that price counts.
The timeline is part of the algorithm
A margin call is not complete when a number appears on screen. The lifecycle is closer to:
- freeze the agreed valuation snapshot;
- value covered trades and net the portfolio under the legal agreement;
- value recognised collateral and in-transit items;
- calculate the call or return amount;
- issue the call in the agreed format;
- validate the counterparty’s calculation;
- agree, partially agree or dispute;
- select eligible collateral if selection is required;
- settle the transfer;
- reconcile settlement status and update collateral balances;
- accrue and settle interest on cash collateral where the agreement provides for it.
ISDA’s collateral-management suggested operational practices emphasise validation of core call data and timely responses. That is important because a correct call delivered after the settlement window can still create liquidity and exposure problems.
Disputes are diagnostic information
Suppose Bank A calls $4.8 million and Bank B calculates $3.9 million. The $0.9 million difference should not immediately be labelled “model error.” A structured dispute tree asks where the states diverged:
- Are the same trades in scope?
- Are trade economics and lifecycle events identical?
- Are both parties using the same valuation time and market-data cut?
- Do curves, volatilities, FX rates or model inputs differ?
- Are collateral balances and pending transfers reconciled?
- Are thresholds, MTA and rounding terms mapped correctly?
- Is the same legal netting set being used?
- Are collateral haircuts and eligibility rules aligned?
The dispute amount is therefore evidence. If trade-level valuations reconcile but the collateral balance does not, the weak link is probably operational state rather than pricing. If collateral agrees but portfolio exposure differs sharply, valuation or trade-population differences become more plausible.
Failure modes and counterexamples
Stale collateral state: yesterday’s settled movement is still marked “in transit,” so the system subtracts it twice.
Wrong sign convention: the same economic exposure is represented with opposite signs by two subsystems, turning a delivery into a return.
Netting-set leakage: trades from two legal agreements are combined because they share the same counterparty name. Economic aggregation has silently overridden legal scope.
Haircut mismatch: one system uses market value while another uses collateral value after haircut.
Rounding before MTA: changing the order of operations can change whether a call is generated at all.
False zero: no call is generated not because exposure is zero, but because equal and opposite errors cancel. A zero output is not proof that the engine is correct.
Verification tests that should be able to disagree with the engine
- Independent recomputation: recalculate exposure, collateral value and transfer amount from frozen inputs outside the production engine.
- State reconciliation: match every settled, pending, failed and substituted collateral movement to the ledger.
- Boundary tests: place the raw call just below, exactly at and just above the MTA and rounding boundaries.
- Sign reversal test: view the same netting set from the counterparty perspective; delivery and return should invert consistently.
- Trade-population checksum: both parties should be able to identify which transactions feed the valuation set.
- Collateral shock test: change a haircut or FX rate and verify the call responds in the expected direction.
A useful falsifier is simple: if the call cannot be reconstructed from an immutable snapshot of trades, market data, agreement terms and collateral state, the process is not sufficiently auditable, even if the counterparty happened to agree with the number.
Connections across Bukit Timah Tutor
This bilateral workflow should be distinguished from central-counterparty margin, where clearing-house rules and default waterfalls define a different institutional mechanism. It connects to ISDA SIMM initial-margin calculations, but VM and IM answer different risk questions. It also links naturally to collateral optimisation once the required amount is known, and to SA-CCR when collateral state feeds regulatory counterparty-exposure measurement.
Current status and update triggers
The BCBS and IOSCO reviewed implementation of the non-centrally cleared derivatives margin framework in December 2025 and reported no material issues requiring changes to the global framework. ISDA continues to update operational practices and jurisdictional documentation; for example, in June 2026 it published an updated Asia-Pacific/Japan-region paragraph for its 2016 English-law VM CSA covering, among other regimes, Singapore. Re-check this article when a relevant jurisdiction changes its uncleared-margin rules, when ISDA updates the governing documentation or operational standards, or when collateral eligibility, haircut or settlement conventions change.
Primary and high-quality references
- ISDA, Collateral Management Suggested Operational Practices.
- Basel Committee and IOSCO, Review of the implementation of margin requirements for non-centrally cleared derivatives, 12 December 2025.
- Basel Committee and IOSCO, Margin requirements for non-centrally cleared derivatives.
- ISDA, 2016 Credit Support Annex for Variation Margin — English law.
Educational boundary: The equations here are generic learning models. Contract wording and regulatory rules govern actual calls, not this article.