Model Risk

Model risk is the possibility of adverse decisions or financial consequences from incorrect, misused, or poorly governed model output.

Model risk is the possibility of adverse financial, operational, customer, compliance, or strategic consequences from decisions based on incorrect, misused, or poorly governed model output. It can arise from flawed design, weak data, coding errors, inappropriate assumptions, use outside the intended scope, inadequate validation, or failure to respond when performance deteriorates.

A model can be mathematically correct and still create model risk if it answers the wrong question, uses unrepresentative data, hides material limitations, or is applied to a different product, population, or market condition than intended.

Key Takeaways

  • Model risk comes from model error and inappropriate model use.
  • Materiality depends on the decision and exposure affected, not only on technical complexity.
  • Validation should be independent enough to challenge conceptual design, implementation, data, performance, and use.
  • Back-testing is important when outcomes can be observed, but it cannot prove performance under conditions absent from the test period.
  • Vendor and machine-learning models still require inventory, documentation, validation, monitoring, and accountable use.
  • Overlays and expert judgment can address known limitations, but they add assumptions and governance requirements of their own.

What Counts as a Model?

Definitions vary by organization and regulatory framework. A model generally transforms input data through quantitative methods, assumptions, or rules to produce estimates, classifications, forecasts, valuations, risk measures, or decision support.

Examples include:

  • credit scoring and probability-of-default estimates
  • fraud and anti-money-laundering monitoring models
  • security and derivative valuation models
  • interest-rate, liquidity, and capital projections
  • expected-credit-loss and reserve models
  • value-at-risk and stress-testing models
  • insurance pricing and reserving models
  • portfolio optimization and asset-allocation tools
  • algorithmic trading signals
  • machine-learning systems used for financial decisions

Spreadsheets, rules engines, scorecards, and expert-judgment processes may or may not fall within a formal model definition. Excluding a tool from the model inventory does not eliminate its risk. The organization still needs controls proportionate to the consequences of error or misuse.

Main Sources of Model Risk

SourceWhat can go wrongExample evidence
Conceptual designTheory, variables, structure, or assumptions do not represent the intended relationshipDevelopment documentation, challenger analysis, sensitivity tests
DataInputs are inaccurate, incomplete, stale, biased, or unrepresentativeData lineage, quality tests, missing-value treatment, population comparison
ImplementationCode, configuration, interfaces, or reports differ from the approved designCode review, parallel run, reconciliation, change records
EstimationParameters are unstable, overfit, or based on an unsuitable sampleOut-of-sample testing, uncertainty ranges, benchmark results
UseOutput is applied outside its approved purpose, population, horizon, or conditionsUse inventory, policy, user procedures, decision records
InterpretationUsers treat an estimate as precise or ignore limitations and uncertaintyTraining, reports, disclosures, override rationale
ChangeProduct, customer, market, regulation, or data-generating process shiftsDrift monitoring, stability analysis, change assessment
AggregationIndividually reasonable outputs interact inconsistently across modelsReconciliation, dependency map, enterprise scenario testing
Third partyVendor logic, data, updates, or limitations are not sufficiently understoodContract, technical documentation, validation access, contingency plan
RiskMain focusRelationship
Model riskIncorrect or misused model outputCan produce poor decisions even when systems run as designed
Operational riskFailed people, processes, systems, third parties, or external eventsCoding, deployment, access, and process failures can create model risk
Market riskLoss from prices, rates, spreads, currencies, or volatilityA market-risk model can understate the underlying market exposure
Credit riskBorrower or counterparty failureA credit model can misestimate default, exposure, or recovery
Data riskUnfit, unavailable, inaccurate, or poorly governed dataData weakness is a major source of model risk
Decision riskPoor judgment or governance around a decisionA sound model can be overridden or interpreted poorly

These categories overlap. Classification is less important than identifying the failure pathway, owner, affected decision, and control response.

Model Risk Management Lifecycle

1. Define Purpose and Materiality

Document the decision, users, products, population, geography, horizon, outputs, dependencies, and consequences of error. A simple model used for a major capital or pricing decision may deserve more control than a complex model used only for exploratory analysis.

2. Develop and Document

Explain the theory, assumptions, variables, data, transformations, estimation method, limitations, uncertainty, and intended use. Development evidence should allow a qualified reviewer to reproduce the logic and challenge alternatives.

3. Implement and Test

Confirm that code, data pipelines, interfaces, calculations, reports, permissions, and downstream uses match the approved design. Reconcile test output and preserve version and change evidence.

4. Independently Validate

Validation should be sufficiently independent from development and use. Common elements include:

  • evaluation of conceptual soundness
  • review of data and implementation
  • ongoing monitoring and process verification
  • benchmarking or challenger analysis
  • outcomes analysis and back-testing where appropriate
  • review of limitations, overrides, and compensating controls

Validation depth should be proportionate to the model’s materiality, complexity, uncertainty, and use.

5. Approve With Conditions

Approval should specify permitted uses, limitations, performance thresholds, monitoring, validation frequency, open findings, overlays, and conditions that require restriction or redevelopment.

6. Monitor Performance and Use

Monitor input drift, output stability, observed outcomes, overrides, exceptions, data quality, user behavior, and changes in market or business conditions. A model can remain technically unchanged while its environment changes materially.

7. Change, Restrict, or Retire

Material changes should trigger review, testing, documentation, and approval. Models with unresolved severe weaknesses may require use restrictions, conservative adjustments, replacement, or retirement.

Worked Example: Credit Model After a Portfolio Shift

Assume a lender developed a default model using seasoned salaried borrowers. The lender later expands rapidly into newly formed small businesses but continues using the same model and cutoffs.

The model may still execute exactly as coded, yet model risk increases because:

  • borrower cash flows and default drivers differ
  • the development sample does not represent the new population
  • limited performance history delays reliable outcome analysis
  • economic sensitivity and documentation quality may differ
  • the original calibration and approval did not cover this use

A sound response could include restricting use, developing a separate segment, applying conservative overlays, increasing manual review, collecting new outcome data, testing sensitivity, and obtaining validation and approval before broader reliance.

The overlay does not eliminate model risk. It needs a documented basis, owner, amount or method, approval, monitoring, and exit criteria.

How to Evaluate a Model

Ask:

  • What decision does the model affect, and how material is that decision?
  • What is the model’s intended use and prohibited use?
  • Which population, product, geography, horizon, and conditions does it cover?
  • Are input data complete, timely, representative, and traceable?
  • Which assumptions drive the result?
  • How sensitive is output to plausible input and parameter changes?
  • Was implementation reconciled with the approved design?
  • What independent validation was completed before use?
  • Which outcomes can be observed, and how quickly?
  • Are performance thresholds tied to action?
  • How are overrides, overlays, exceptions, and limitations governed?
  • What happens if the model or a vendor becomes unavailable?

Back-Testing, Benchmarking, and Stress Testing

These methods answer different questions:

MethodQuestionLimitation
Back-testingDid predictions align with subsequently observed outcomes?Outcomes may be delayed, sparse, or influenced by intervention
BenchmarkingHow does output compare with an alternative method or external reference?The benchmark can also be weak or differently scoped
Sensitivity analysisWhich inputs and assumptions drive the output?Small changes may not represent structural breaks
Stress testingHow does the model or decision behave under adverse assumptions?Results depend on scenario design and model behavior outside observed data
Challenger modelWould another defensible approach materially change conclusions?Multiple models can share data and conceptual weaknesses

Passing one test is not proof of overall validity. Evidence should be considered together and in the context of intended use.

Machine Learning and Third-Party Models

Complexity can make development and validation harder, but model risk is not limited to opaque methods. A simple rule can be materially wrong, while a complex model can be controlled if its use, data, behavior, limitations, and outcomes are understood.

For machine-learning and vendor models, reviewers may need to address:

  • training and evaluation populations
  • feature stability and permissible use
  • explainability appropriate to the decision
  • bias and differential outcomes
  • data and concept drift
  • version changes and vendor dependencies
  • access to documentation and validation evidence
  • fallback procedures and exit rights

Limited access to vendor intellectual property does not remove accountability for the decision.

Common Mistakes

  • Treating model validation as a one-time approval.
  • Equating good historical fit with reliable future use.
  • Allowing use outside the approved population or purpose.
  • Ignoring judgmental overlays because they are outside the coded model.
  • Validating calculations while overlooking data lineage and downstream reports.
  • Using an average performance measure that hides weak segments or tail outcomes.
  • Assuming a vendor certification replaces internal review.
  • Tracking findings without deadlines, compensating controls, or use restrictions.
  • Calling a tool “not a model” to avoid proportionate governance.

Official Sources

The Federal Reserve’s April 2026 guidance superseded SR 11-7 for its stated supervisory scope and identifies the banking organizations to which it is most relevant. These banking materials are useful reference points, not universal requirements for every model, company, investor, or jurisdiction.

  • Operational Risk: Process, system, implementation, data, or human failures that can create or amplify model-related loss.
  • Fraud Detection: A common model use where changing behavior, incomplete labels, and false positives require careful governance.
  • Corporate Failure Prediction: A model family whose event definitions, samples, calibration, and data timing affect reliability.
  • Credit Migration Rate: A transition measure whose use depends on stable grades, data lineage, horizon, and withdrawal treatment.
  • Reputational Risk: A possible consequence when model-driven outcomes undermine stakeholder trust.

FAQs

What is model risk in simple terms?

Model risk is the possibility that a model is wrong, is used incorrectly, or is governed poorly enough that its output contributes to an adverse decision or financial consequence.

Can a statistically accurate model still create model risk?

Yes. It may be used for the wrong population, purpose, horizon, or market condition, or users may ignore uncertainty and limitations.

Does validation eliminate model risk?

No. Validation can identify weaknesses and support controlled use, but data, behavior, products, markets, and model use continue to change.

Educational Use

This article provides general financial education. It is not personalized banking, investment, credit, accounting, technology, legal, regulatory, statistical, or model-validation advice.

Browse Risk Management