Case study · XVA · In production
XVA in production since 2018 at a European financial institution
How a Monte Carlo XVA platform went live on-premises and has stayed in daily use for years. The client is not named, and no figures are given.
Last reviewed
At a glance
- Client: a European financial institution with an OTC derivatives portfolio.
- Scope: CVA, DVA, FVA and LVA from Monte Carlo exposure simulation (G2++ for rates, Heston for stochastic volatility), with allocation down to the trade.
- Deployment: on-premises, inside the client's own infrastructure.
- Status: in production since 2018 and in daily use, run as a scheduled workflow.
The challenge
The institution needed valuation adjustments on its derivatives portfolio that it could compute itself, explain line by line and defend to auditors. Three constraints shaped the project:
- Transparency. The risk team had to see why a number moved: which counterparty, which netting set, which trade, which market input.
- Control of data. Trade and market data could not leave the institution's infrastructure, which ruled out a hosted service.
- Model quality. Exposures had to come from a proper simulation of future market states, not from static add-ons, so that every adjustment would rest on the same exposure profiles.
Scope
What runs in production today, in one table.
| Area | In production |
|---|---|
| Since | 2018, continuously, with platform upgrades delivered by Everix |
| Portfolio | The institution's OTC derivatives book |
| Adjustments | CVA, DVA, FVA and LVA per counterparty and netting set |
| Exposure engine | Monte Carlo simulation of market risk factors, repricing every trade on every path and date; EPE, ENE and PFE profiles |
| Models | G2++ (two-factor Gaussian) for interest rates; Heston for stochastic volatility |
| Allocation | XVA allocated to each trade, so contributions can be analysed and explained |
| Netting | Exposure aggregated by netting set on every path |
| Integration | Automated loading of trades and market data from the client's existing systems |
| Deployment | On-premises, operated within the client's infrastructure and security perimeter |
| What runs daily | Load trades and market data, simulate, compute every adjustment, allocate to the trade and publish the reports, as one scheduled workflow with no manual steps |
Since
- In production
- 2018, continuously, with platform upgrades delivered by Everix
Portfolio
- In production
- The institution's OTC derivatives book
Adjustments
- In production
- CVA, DVA, FVA and LVA per counterparty and netting set
Exposure engine
- In production
- Monte Carlo simulation of market risk factors, repricing every trade on every path and date; EPE, ENE and PFE profiles
Models
- In production
- G2++ (two-factor Gaussian) for interest rates; Heston for stochastic volatility
Allocation
- In production
- XVA allocated to each trade, so contributions can be analysed and explained
Netting
- In production
- Exposure aggregated by netting set on every path
Integration
- In production
- Automated loading of trades and market data from the client's existing systems
Deployment
- In production
- On-premises, operated within the client's infrastructure and security perimeter
What runs daily
- In production
- Load trades and market data, simulate, compute every adjustment, allocate to the trade and publish the reports, as one scheduled workflow with no manual steps
Approach
- Data first. Trade, netting and collateral data were mapped into the Everix trade model before any model work, with reconciliation reports so the client could see what was loaded and what was rejected.
- Models with the risk team. Model choices and calibration were reviewed with the client's quantitative team, with documentation they could hand to validation and audit.
- Parallel runs. Results were run side by side with the client's existing figures until differences were understood and agreed with the risk team.
- Automated production. Loading, simulation and reporting run as a scheduled workflow, so a production run needs no manual steps.
- Long-term maintenance. Since go-live, Everix has delivered platform upgrades and model maintenance, checking results run against run before each release so that nothing moves without an explanation.
Outcome
- The platform has been in production continuously since 2018.
- The risk team computes CVA, DVA, FVA and LVA in-house and explains changes down to the trade.
- Data stays inside the institution's infrastructure.
What we learned
- Explainability beats speed. The questions that matter in production are “why did it move?” and “which trades drive it?”, so per-trade allocation and drill-down earned their keep from the first month.
- Reconciliation is part of the product. Clear reports on what data was loaded, and what was not, let the team fix data issues before a run reaches production.
- Upgrades must be boring. A long-running platform needs a release process that proves results are unchanged unless a change was intended.
The capabilities behind it
- In production
Monte Carlo XVA (CVA, DVA, FVA, LVA) with per-trade allocation
In production since 2018. See the XVA module. - In production
On-premises deployment
Runs inside the client's infrastructure. - Available
SA-CCR, ISDA SIMM and IFR on the same trade model
The same engine and trade model now carry the regulatory modules. See counterparty risk core. - Available
AI-agent access to results (MCP)
Read-only, scoped and audited. See the MCP docs.
Published with the client's identity withheld. We do not disclose client names, portfolio sizes, trade counts or any calculated values.
Related on everix.io