A margin model is a forecast. It is the counterparty's best estimate of how much capital must be posted today to absorb the worst plausible loss tomorrow. Like any forecast, it can be conservative or aggressive, and like any forecast it can be quietly recalibrated when conditions change. The interesting question for an institutional desk is not whether a particular counterparty's margin model is conservative on a normal Tuesday. It is what happens to that model on the Tuesday after a Friday close that gapped six standard deviations.
Margin models that scale up faster than the underlying risk are pro-cyclical. They raise the capital requirement on a position precisely when the desk has the least free capital to post. They are the financial equivalent of a fire alarm that locks the exit door. They are also, unfortunately, the default behaviour of most margin engines.
This piece walks through what pro-cyclical margin actually looks like in the stress windows of 2022, 2023 and 2024, what the alternative looks like, and what an institutional desk should ask of a counterparty before the next stress window arrives.
What pro-cyclicality looks like in practice
In the March 2020 dash-for-cash, the most-liquid G10 FX pairs saw realised one-day volatility multiply by three to five over a two-week window. Many margin models, calibrated to a trailing 90-day volatility window, took the new reading and immediately scaled up their margin requirements proportionally. A position that required $1M of collateral on March 1 required $4M on March 15.
Some desks could meet the call. Many could not — not because the underlying P&L was lost, but because the cash to top up margin was tied up elsewhere in the same stress. The desks that were forced to liquidate did so into a market that was already disordered, realising a worst-case execution cost and locking in losses that a slightly more patient model would have absorbed without forcing the sale.
The 2022 sterling crisis, the 2023 regional-bank window and the 2024 yen carry-unwind each produced milder but qualitatively identical episodes. In each case, counterparties whose margin models scaled aggressively into the stress contributed to the disorder. Counterparties whose models scaled more slowly absorbed the stress without amplifying it.
The mechanics: VaR, expected shortfall, and the look-back window
Most institutional margin models are built around either a one-day 99% Value-at-Risk number or an Expected Shortfall variant at a similar confidence. The free parameter, in either case, is the look-back window over which the historical or simulated returns are evaluated.
A short look-back window — 30, 60 or 90 days — responds quickly to recent volatility, which means the margin requirement tracks current conditions closely. The cost is pro-cyclicality: a stress window pushes the margin requirement up sharply, exactly when posting capital is hardest. A long look-back window — one year, two years, or longer — smooths the response and provides counter-cyclical capacity. The cost is over-collateralisation during quiet periods, which makes the counterparty commercially uncompetitive on a normal Tuesday.
The compromise that some sophisticated counterparties run is a hybrid: a long-window baseline that sets the floor, plus a short-window add-on that activates only when the short-window estimate exceeds the long-window estimate by a configured threshold. The hybrid maintains baseline efficiency in normal times while damping the worst of the pro-cyclical response in stress. Few counterparties run it, because doing so requires sustained capital headroom in quiet periods, which conflicts with shareholder pressure for return on equity.
Initial margin, variation margin, and the cliff
Pro-cyclicality also lives at the boundary between initial and variation margin. Variation margin tracks daily mark-to-market and is operationally routine. Initial margin sits on top and absorbs the worst plausible one-day loss between the desk's failure and the counterparty's ability to close out the position.
The cliff arises because initial margin is, in many counterparty agreements, a step function of credit-risk indicators. If the desk's CDS spread breaches a threshold, or its rating drops a notch, or its reported drawdown exceeds an agreed limit, the initial margin requirement jumps — often by 50% or more. The cliff is itself pro-cyclical: it activates in exactly the conditions where the desk has the least flexibility to absorb a step change.
Some counterparties have moved away from step-function cliffs and toward continuous margin schedules indexed to a smoothed credit-risk metric. The continuous version is operationally less convenient but materially less amplifying in stress. It is also the version that an institutional desk should require from any counterparty whose stress-cycle behaviour matters.
What an institutional desk should ask
Before signing an ISA or equivalent counterparty agreement, the desk should ask the counterparty four specific questions about its margin model. The answers are diagnostic of how the counterparty will behave in a stress.
First: what is the look-back window for the variance estimate? A counterparty using a 60-day window will scale margin aggressively into the next stress. A counterparty using a 250-day window or a hybrid will scale more slowly. Neither is wrong; the desk should know which one it is exposed to.
Second: is the credit-risk overlay continuous or step-function? A step function is a hidden gap risk in the counterparty agreement; a continuous overlay is operationally smoother but commercially more honest.
Third: under what conditions does the counterparty have unilateral discretion to raise initial margin beyond the schedule? Many ISAs contain a clause permitting the counterparty to require additional margin "as it deems necessary in light of prevailing market conditions." In quiet times this is dormant; in a stress it is the cliff above all cliffs. The desk should know what governance applies to its activation.
Fourth: what is the counterparty's own funding cost in the relevant currencies, and how is it passed through? If the counterparty is itself running on short-term wholesale funding that becomes expensive in a stress, the desk's funding charge will scale with the counterparty's stress, even if the desk's own position is unchanged.
How Drovix runs its margin model
Drovix's institutional margin model is a hybrid 250-day Expected Shortfall baseline with a 60-day stress overlay that activates only when the short-window estimate exceeds the long-window estimate by more than 30%. The counterparty's CDS spread and rating do not enter the margin formula as step functions; they enter as continuous inputs to a smoothed credit-risk multiplier that is reviewed quarterly and disclosed to the counterparty.
The counterparty's funding charge in each currency is set against a disclosed reference curve, with the spread negotiated bilaterally in the ISA and held stable through the term unless there is a documented change in the counterparty's market funding cost. Discretionary authority to raise margin beyond the schedule sits with the credit committee and requires written notice and disclosed rationale, with the counterparty's right to a 48-hour response window for unwinds.
None of this prevents a real stress from producing a real margin call. What it does is keep the call's size within the desk's planning envelope, and keep the counterparty's behaviour predictable across the cycle. That predictability is what makes a counterparty usable as a strategic execution venue rather than a tactical one.
The downstream consequence of pro-cyclical margin on the desk's effective spread shows up most starkly in stress windows; the anatomy of effective spread covers that compounding directly. The structural question of how a desk diversifies counterparty exposure across multiple margin regimes is in counterparty concentration risk.
Methodology and data
Procyclicality is quantified as the percentage change in initial margin required by tier-1 prime brokers during the five highest-realised-vol days of each quarter from Q1-2024 through Q1-2026. Margin schedules are normalised to a constant-notional EUR/USD position; cross-margin offsets are excluded to make brokers comparable. Sample: nine tier-1 PBs disclosed pseudonymously. Stress windows are identified ex-post by 1-day realised volatility exceeding the 95th percentile of the prior 250 trading days.
Limitations and scope
Margin schedules outside the EUR/USD-USD/JPY-XAU/USD core are not directly comparable across PBs because the haircut models differ structurally. The study captures schedule changes that PBs announced; it does not capture discretionary margin calls placed by individual relationship managers. Procyclicality measured ex-post is not predictive of any individual PB's behaviour in the next stress window.
Further reading
→ Balance-Sheet Cost of Leverage — Why a wider PB margin schedule reprices every position you hold. See /blog/balance-sheet-cost-of-leverage.
→ Counterparty Concentration — 2026 Diligence — Diligence questions a broker should answer about its PB exposure. See /blog/counterparty-concentration-2026-diligence.
Drovix Research is the in-house institutional desk of Drovix MU Ltd, regulated by the Financial Services Commission of Mauritius. All notes are informational only and do not constitute investment advice, a solicitation, or a recommendation to transact in any financial instrument.
Analyst Desk
Drovix Research Desk
Institutional Research
Drovix Research Desk publishes institutional-grade analysis covering macro events, cross-asset correlations, and execution insights for professional market participants.