Simply closer
to financial markets

Example: breakdown of model validation for FX pricing models

A concrete view of an independent validation — here for FX pricing models — from theoretical soundness through to live-portfolio behaviour.

E

Theoretical soundness & documentation review

Model assumptions, derivation and limitations, against the documentation.

E

Product-pricing benchmarks

Model × product comparison — first-generation exotics and TARFs across types, with model-dependency analysis.

E

Numerical analysis

PDE convergence and PDE-vs-Monte-Carlo agreement; grids, stable schemes and smoothness of Greeks.

E

Model reserves & prudent valuation

Fair-value and prudent-value adjustments (FVA / AVA).

E

Front-office model replication

Independent re-implementation / challenger models to characterise a vendor or in-house model.

E

Market data & calibration benchmarking

Implied and local volatilities, calibration quality and stability.

E

Payout validation

Payoff handling, event scheduling, fixings and exercise; PDE and Monte-Carlo engines validated independently.

E

Live-portfolio & scenario tests.

Behaviour across stress and market scenarios on a real portfolio.

E

Independent price verification

IPV using consensus-pricing services, with mixing-factor recalibration where relevant.

E

Compliance & governance review

Roles, tiering and policy alignment, woven into the validation record.

Calibration check — FX Local Volatility model

One concrete step from the independent and highly automated validation of the FX Local Volatility model: the calibrated Local-Volatility surface (Lv) is checked against an independent challenger calibration (SmileEu) across 5Δ × 9 tenors. Both reproduce every raw market pillar to numerical noise — the maximum implied-vol deviation is 1.0 bp against a 4.0 bp threshold, so the calibration check passes across the full surface, including the short end.

Top: implied-vol smile per tenor (1W–2Y) — raw 5Δ market pillars (green), the independent challenger SmileEu and the model’s Lv overlay in close agreement. Bottom: heatmap of the residual (Lv − SmileEu) in implied-vol basis points; all cells stay well inside the 4 bp tolerance, with a maximum of 1.0 bp.

Address

Kaiserstr. 50
60329 Frankfurt am Main, Germany

Email

modelriskmanagement@mathfinance.com