Customers / Case study

A margin dispute that took four days now takes an afternoon

Major UK bank

5 minute read SIMM, FRTB-SA

A major UK bank exchanges initial margin with 31 counterparties. When two SIMM numbers differed, finding out why meant two teams, two sets of spreadsheets and a week of email. Now the file says where they differ.

The result in three figures

  • 5.8 h

    Median time to settle a margin dispute

    Before: 4.2 days

  • −71%

    Disputed margin calls over two quarters

  • 31

    Counterparties compared line by line

The situation

The bank exchanges initial margin on uncleared derivatives with 31 counterparties. Every night its margin team computes SIMM from the sensitivities that the risk system sends them, and every counterparty does the same. Most nights the numbers agree to the cent. On about one night in eight they do not, and a call is disputed.

Settling a dispute meant finding the trade, risk factor or bucket where two implementations had parted. Neither the vendor system nor the in-house reconciliation could explain its own figure beyond the total, so the work was done by hand, across two sets of spreadsheets, by people on two sides who each believed the other was wrong. The median dispute took four days to close. The longest took five weeks, and was resolved when someone noticed that the two sides had used different versions of a static data table.

What changed

The bank ran GraphAtlas SIMM beside its production engine for eleven weeks. The engine reads the same risk data file, works the margin through the risk classes and buckets down to the trade, and attributes every part of it exactly, so that the lines add up to the total with nothing left over. When a counterparty sends its own file, the comparison ranks the lines where the two differ by the margin that they explain.

In the second week the parallel run showed a difference of 0.4% in equity vega margin that no one had noticed. It traced to the bucket given to a class of depositary receipts, in a static data table that the production system had last updated eighteen months earlier. The correction reduced the bank’s own margin by a little and removed the most frequent cause of its disputes.

The result

Since go-live in March the median dispute closes in 5.8 hours. Disputed calls fell by 71% over the first two quarters, and the margin team no longer keeps spreadsheets for the work. Internal audit accepted the receipt of each night’s run as the record of how the number was made, and asks for no other.

In the second half of the year the bank is extending the same engine to its FRTB-SA run, so that the desk heads who see their margin in these terms see their capital in them too.

When a counterparty’s number differs from ours, we no longer argue about whose model is right. We open the same file and the difference is already ranked by line.

Head of Market Risk Methodology Major UK bank

What they run

The engines behind it

This customer runs 2 engines from GraphAtlas, on the same kernel. Each page shows the calculation, the clauses it covers and the evidence for it.

The essay behind it

Numerics

Where a capital number comes from

A capital charge is a square root, so it does not add up by position. How the engine splits it exactly, and what the page does when it rounds the pieces.

Marcus Teller and Tomasz Wierzbicki6 min read

The next step

Bring us
one book

A scoping call, a sample book and a parallel run on your data. Most teams have the first reconciled number inside two weeks, and a receipt for it on the day.

Book a demo Work out a price