The risk that financial, valuation, forecasting, or decision-support models produce incorrect outputs or lead to inappropriate decisions because of coding errors, mis‑specification, flawed assumptions, incorrect inputs, or misuse in context of their intended purpose.

Principle

Model outputs are conditional on model structure and inputs; therefore errors or inappropriate assumptions propagate into decisions and exposures unless identified and adjusted.

Demonstration

Situation: A trading desk uses a volatility model to set daily limits. Recognition: Backtesting shows consistent underprediction of tail losses after a regime shift. Action: The desk suspends reliance on the model, applies conservative overlays and re‑calibrates. Consequence: Immediate exposure from previously accepted positions is re‑priced or reduced, preventing further accumulation of unrecognized tail risk.

Misapplication

Mistaken interpretation: Treating good historical fit as proof the model will perform under new market regimes. Semantic error: Conflating calibration performance (in‑sample fit) with structural validity under changed conditions.

Consequence

When model error is unrecognized, decisions (pricing, hedging, capital allocation) can systematically misstate risk or value, producing direct financial losses, incorrect strategic choices, or regulatory capital shortfalls via a causal chain: flawed model → mistaken decision → realized loss or inadequate reserves.

Reversal

The principle weakens when model outputs are used only as advisory inputs within decision processes that embed independent safeguards (stress tests, conservative overlays, limit checks) or when the model maps a purely descriptive, non‑binding metric; in those cases model errors have limited operational consequence.

Boundary

Clearly within: A mis‑specified credit scoring model that underestimates default probabilities. Boundary case: A model with known parameter uncertainty but explicitly accompanied by stress scenarios and governance—risk exists but is mitigated. Clearly outside: an IT outage that prevents model execution but does not reflect a model's logical error.

Semantic Tension

Accuracy and complexity ↔ Interpretability and robustness: more complex models may fit data better but can be harder to validate and more fragile under regime change.

Synthesis

Model risk is not just about bugs or wrong numbers; it is the systemic dependence of decisions on conditional assumptions—managing it requires treating models as fallible information sources and embedding procedures that expose, stress and limit reliance on them.