Model risk & governance
Turn assumptions, limitations and residuals into controls.
01Separate model, parameter and implementation risk
02Design challenger and benchmark tests
03Define monitoring and escalation thresholds
Define the exposure before compressing it into a metric.
A model is a controlled approximation: governance begins where mathematical validity ends.
Documentation is an executable contract.
Independent validation targets failure modes.
Monitoring connects residuals to action.
Fix portfolio, scenarios, horizon, and legal terms.
Prices, limits and capital are only defensible when model ownership, evidence and fallback actions are explicit.
pricing libraries
risk engines
calibration services
Every metric states owner, threshold, observation window and escalation path.
Aggregate with an explicit measure and convention.
Validation residual
Residuals must be segmented by product and regime, not hidden in a global average.
Traffic-light control
A breach is useful only when it triggers a documented response.
Short derivation
From information set to computable quantity
Each line states the information, measure and unit before manipulating the expression.
- 01
Define intended use
State products, markets, users and prohibited extrapolations.
- 02
Map assumptions to tests
Each material approximation receives an invariant, benchmark or stress.
- 03
Set thresholds
Use economic materiality and numerical evidence, not arbitrary round numbers.
- 04
Close the control loop
Assign owner, cadence, escalation and fallback for every breach.
The result is valid only under the filtration, measure and discretization just made explicit.
Inputs
εmodel: benchmark residualT: escalation threshold
Assumptions and limits
- Governance cannot eliminate judgement or unknown unknowns.
- Overly broad metrics can hide localized failure.
Reconcile valuation, risk, and model limitations.
Maintain an assumption-to-test matrix with versioned evidence, challengers, thresholds, owners and fallback procedures.
Monitor parameter stability and calibration residuals by regime; do not treat refits as automatic remediation.
06PYTHON IMPLEMENTATIONOpen the implementation and checks.
- Typed domain validation
- Deterministic seeded computation
- Readout plus invariant
Model risk & governance
Reproduce the governing quantity, then challenge it with an invariant.
import numpy as np def varepsilonivimodel(x: np.ndarray) -> float: x = np.asarray(x, dtype=float) assert np.isfinite(x).all() return float(np.mean(x)) sample = np.array([0.8, 1.0, 1.2])value = varepsilonivimodel(sample)assert sample.min() <= value <= sample.max()print(f"value={value:.6f}")Move the state. Challenge the equation.
Model risk & governance
Change scale, volatility, collateral and confidence. Exposure, tail and adjustment metrics respond from one synthetic portfolio.
Synthetic scenario losses with the selected historical VaR threshold.
- scenario loss
- VaR
Use Left/Right or Up/Down arrows to inspect values; Home and End jump to the bounds.
Scenario → distribution → decision
Legal terms and model state enter before the summary metric and its governance action.
18.00% volSynthetic market scenarios
0.35mUnsecured exposure boundary
2.55mDecision metric with explicit convention
Turn exposure into a controlled decision.
“Green status means evidence met a threshold, not that the model became true.”
model inventory
validation evidence
production telemetry
Monitor parameter stability and calibration residuals by regime; do not treat refits as automatic remediation.
RISKscope creep
stale calibration
implementation drift
- Validate market state and timestamp
- Recompute the baseline
- Run a controlled perturbation
- Explain P&L and residuals
Production failure modes
- Silent convention or measure changes
- Unstable numerics hidden by plausible prices
09MACRO CONNECTIONOpen the transmission channel.
Transmission from state to valuation
The causal chain separates the economic shock from the modelling response.
transmitscreates decision dependence
transmitstests known failure modes
outputlimits and escalates residual risk
10COMMON PITFALLSOpen the failure checklist.
Equating a successful calibration with validation
Monitoring averages that hide tail breaches
11SOURCES / FURTHER READINGOpen sources and continue the track.
Measure theory, simulation and computational-finance lectures
The lesson uses original prose and a fresh typed implementation; the linked material is a research map, not copied product code.
- Source
- Computational Finance Course
- Author
- L. A. Grzelak
- Ref
- main