AAD enablement AAD-enable your engine.
We build XAD, commercially available AAD since 2014, into your pricing and risk engine and keep it current: every sensitivity of the book for as little as 2 times the cost of one valuation with XAD's tape and, with code generation, the whole run, every sensitivity included, up to 3 times faster than one plain valuation of the original code, in your own library, under your governance.
Even after making a valuation ten times faster, ten thousand sensitivities with bumping still cost a thousand original valuations. XAD changes the arithmetic: every sensitivity for as little as 2 times the cost of one valuation with XAD's tape. With code generation, the whole run, every sensitivity included, is up to 3 times faster than one plain valuation of the original code.
Also with AAD
QuantLib with AAD
QuantLib built with XAD: all sensitivities of a pricing in one adjoint pass, with QuantLib's code unchanged and tested every day against its master branch. Open source under the AGPL, with a commercial licence for use in your organisation.
QuantLib with AADAAD training
The AAD course taken by quant teams at leading banks, taught by the team that built XAD, with a hands-on workshop on your own code.
AAD trainingAlready have AAD?
Xcelerit Verified checks every sensitivity of your own AAD implementation against bumping, independently, with a report your validation team can sign off.
Xcelerit VerifiedWhere AAD pays
- FRTB sensitivitiesDelta and vega for the sensitivities-based method, every day, without one revaluation per risk factor; the curvature shocks run in the time freed.Case: FRTB standardised approach sensitivities
- SA-CVACVA sensitivities for capital, overnight on the grid you already have.Case: SA-CVA capital sensitivities
- XVA and hedgingFull XVA sensitivities for hedging and pre-deal checks.Case: whole-book XVA sensitivities
- SIMM and MVAMargin sensitivities for every netting set, and forward SIMM for MVA.Case: MVA with forward SIMM
- CalibrationModel calibration with exact gradients instead of finite differences.Case: daily model calibration
Reported by clients
AAD · XVA
The whole-book sensitivity run from days to about an hour
Reported by the client · Global investment bank
Whole-book XVA sensitivities with AAD
Every first-order XVA sensitivity across 8,000 netting sets from one AAD pass of the bank's own code: the full sensitivity run from days to about an hour on the existing grid.
Full case studyAAD · SA-CVA
The capital run from over two days of grid time to about an hour
Reported by the client · Tier-one bank
SA-CVA capital sensitivities with AAD
About 300 MAR50 market-risk sensitivities, with vega, and the counterparty credit spreads from one AAD pass: the capital run from over two days of grid time to about an hour.
Full case studyAAD · FRTB SA
Sensitivity compute from about 8 hours to under 1, ahead of go-live
Reported by the client · European bank
FRTB standardised approach sensitivities ahead of go-live
Delta and vega for the sensitivities-based method from one AAD pass of the bank's own pricers, curvature on faster revaluations: compute from about 8 hours to under 1 on the same grid, ahead of go-live.
Full case study“The morning SLV calibration across 300 underlyings went from three hours to 25 minutes, and we get vega to each quote through the calibration instead of recalibrating for every bump.”Head of Equity Derivatives Quantitative Analytics, Tier-one bank
“Forward SIMM at every margin date used to be something we approximated. Now it is 14 billion deltas a run, and the MVA of a new trade comes back in about a minute.”Head of Margin Analytics, Major bank
How it works
Minor, non-disruptive changes. AAD with XAD gives every sensitivity for as little as 2 times the cost of one valuation with XAD's tape instead of one bumped revaluation per risk factor. With code generation, the whole run, every sensitivity included, is up to 3 times faster than one plain valuation of the original code.
Speed-up estimate
One calculation, measured on your hardware with your team. Written findings, the gains available, and where AAD and code generation with XAD would pay back. Two weeks, free, with a fixed quote for the build.
- Not ready to share code? The first step needs none: a short call with one of our engineers, and the sample report.
- You send a representative kernel, or one pricer with one test case: not production code, and never production data.
- By default it runs inside your bank, on your hardware: our engineer works with your team on a call, and nothing leaves the bank.
- The two weeks start at the first working session, once the kernel builds on your machine. Before that, only a mutual NDA.
- Where the time and the cost go
- A quantified roadmap and a fixed quote for the build
Build
A short, fixed-scope build inside your code and your environment, with first results in weeks, not quarters, every price checked against the original results and every sensitivity against your bumped numbers. The code is yours.
- AAD enablement: XAD in your pricers and risk engine, plus tooling that speeds up the port
- Your own QuantLib code: extensions, custom products and in-house pricers
XAD subscription
Every build continues under an annual XAD subscription, so the speed and accuracy survive the next library release, the next compiler and the next team.
- XAD Commercial (per seat or per site) or Enterprise (per site) with every build, annual
- Updates, fixes and support included
- The XAD validation pack: edge-case, stress and performance-regression suites, reconciliation against QuantLib and real pricers, and a validation report for your model-risk team on request
- The code we delivered and its verification suite, re-run and kept current through every compiler and library release
- Enterprise adds code generation, a named support engineer and service levels
Code written for you is yours. XAD runs inside it under the subscription.
Request thisEvery sensitivity, checked against bumping on your books
Model validation doesn't take speed on trust, and neither do we. Every build ships with a verification suite that runs on your own books.
- AAD against bumpingEvery sensitivity checked against your bumped numbers, risk factor by risk factor, within tolerances agreed with your validation team.
- Golden booksReference prices and sensitivities for a set of your own trades, frozen and re-run on every change.
- Known answersLimiting cases and identities where the right answer is known in advance, such as put-call parity.
- Tracked tolerancesAny drift in a price or a sensitivity is caught and explained before it reaches production.
You get the reconciliation report your model validation team signs off: the evidence model risk reviews ask for under SR 26-2 and PRA SS1/23. Under the XAD subscription the suite runs again on every XAD release, compiler and library upgrade, so the numbers stay checked as well as fast. XAD itself comes with its own evidence: the XAD validation pack, with a validation report on request.
How a build runs
What you own, what is licensed
Related
Start with a speed‑up estimate
Your code, measured free in two weeks on your hardware or a cloud equivalent, with a fixed quote for the build.
Free · Measured on your hardware or a cloud equivalent · A fixed quote for the build · NDA ready