Module 1
The one question every margin model answers
Every margin methodology — Reg T, portfolio margin, SPAN, a prime broker's house model — is an answer to a single question: if this client defaults right now, and I liquidate their book over the next N days, how much do I lose?
That decomposes into three pieces, and every methodology in this track is a different way of estimating them:
- How far can the position move before I'm out — the risk measure.
- How long am I exposed while unwinding — the liquidation horizon (also called the margin period of risk, MPOR).
- What's wrong with my estimate of the first two — the add-ons.
Hold onto that third bullet. The base risk number is commoditized — everyone can shock a grid. Add-ons are where margin desks differentiate and where the interesting engineering lives (Module 4).
The simplest possible answer is rules-based margin: Regulation T. Fixed percentages, position by position — 50% initial on equities, 25% maintenance. No portfolio effects at all: a long call and a short call on the same underlying are margined separately unless they fit a recognized strategy template. It's the regulatory floor for retail-adjacent accounts, and it's punitive precisely because it is blind to hedging.
Everything else in this track exists because Reg T over-margins hedged books and under-margins concentrated ones. Try it yourself:
🎛 Reg T vs risk-based lab
Portfolio: long 100 shares at $100 — add protective puts and watch the two methodologies diverge
Portfolio P&L repriced at each spot shock (the risk-based grid)
Worst node: -$629 at -15% — with puts on, the loss flattens out below the strike because the hedge pays in the same scenario.
Reg T (rules-based)
$5,190
50% of stock + full premium — blind to the hedge
Risk-based (scenario grid)
$819
Worst-case loss across the grid + premium
Reg T margins each position separately, so the put adds to the requirement even though it removes risk. The risk-based grid reprices the whole portfolio at every node, so the hedge benefit is automatic. Set puts to zero and the two nearly agree; add the hedge and they diverge sharply. Educational tool — not investment advice.
One architecture, many names
A fact that surprises people meeting this material for the first time: ISDA SIMM and the FRTB standardized approach are the same architecture.SIMM was derived directly from FRTB's sensitivities-based approach — bucket the risk factors, collect delta/vega/curvature sensitivities, apply prescribed risk weights, aggregate with prescribed correlations within and across buckets. If you understand one, you've implemented the skeleton of the other: the model that margins uncleared OTC swaps is a rebadged bank capital model.
Both have live calculators in the Fixed Income track: SIMM and FRTB.
Things to try
- • Set puts to 0 — Reg T and the risk-based grid roughly agree on naked long stock.
- • Add 1 put at the 95 strike — Reg T goes up (you paid premium) while risk-based collapses. That gap is the entire case for portfolio margin.
- • Move the strike to 100 — a tighter hedge, an even smaller worst-case loss.
Test yourself
Why doesn't margin aggregate additively — why isn't the margin on a portfolio the sum of the margins on its positions? (Hint: worst-case-loss computed on the aggregate is not the sum of worst cases, for the same reason VaR isn't subadditive in general — see the VaR module.)