Define what a model is, then characterise how model risk can come to affect a financial institution.
The idea
A model is a quantitative method, system or approach that applies statistical, economic, financial or mathematical theories and assumptions to turn input data into quantitative estimates. It has three parts. The inputs are data and assumptions. A processing component does the calculation. A reporting component translates results into something decision makers use. Model risk is the potential for adverse consequences, such as financial loss, poor decisions or reputational damage, from decisions based on incorrect or misused model output. It arises from two broad sources. Fundamental errors include flawed theory, wrong assumptions, bad data, coding mistakes and assumptions that stop holding. Inappropriate use means applying a sound model outside the purpose or domain it was built for. Severity grows with how widely a model is used and how much exposure rests on it. Exam point: a perfect model used for the wrong purpose still creates model risk.
A practice question
Under supervisory guidance on model risk such as SR 11-7, which description best defines a model?
- A quantitative method applying theory and assumptions to turn input data into estimates. - correct
- Any spreadsheet, database or computer program used within a bank to store, move or display data, whether or not any theory is applied.Why not: Equates a model with any software tool; a model must apply theory and assumptions to produce quantitative estimates.
- A written set of qualitative policies and limits that guides how traders take risk.Why not: Describes risk policy or governance, which is not quantitative processing of data into estimates.
- A pricing or VaR engine on a trading system, which excludes credit scoring and capital models.Why not: Restricts the definition to market risk and pricing; credit, capital and other quantitative estimators are models too.
A model has inputs, a processing component built on theory and assumptions, and a reporting component. It is defined by function, not by the technology or the business line.
A harder one
A bank's market risk model has a correct calculation engine and correct input data. However, the final VaR report aggregates desk figures by simple addition in the wrong currency unit, and senior managers set limits using that report. Which component of the model has failed?
- The input component, because the data and the assumptions fed into the calculation were faulty.Why not: The stem says inputs were correct; the failure occurs after the calculation.
- The processing component, because the calculation engine produced wrong estimates.Why not: The calculation engine is stated to be correct; the error arises when results are aggregated and presented.
- The calibration step, because the parameters were estimated on too short a sample.Why not: Calibration is part of input and processing; nothing in the stem concerns the estimation sample.
- The reporting component. - correct
A model comprises inputs, processing and reporting. Errors in reporting can cause as much harm as errors in calculation because decisions rest on what is reported.
The hardest tier
Model A produces a credit exposure estimate that feeds model B, which calculates capital, and model C, which sets limits. Model A has a small but systematic error. Which statement best describes the resulting model risk?
- It is reduced, because the errors that arise in separate downstream models tend to offset one another through diversification.Why not: Assumes diversification across models; a common upstream error is the same in every downstream model and does not diversify.
- It depends only on how often model A is wrong, not on the size of the exposures relying on it.Why not: Ignores materiality; severity depends on how widely the model is used and the exposure that rests on it.
- It is amplified, because the error passes through every downstream model and decision. - correct
- It is contained, because models B and C each independently validate the output they receive from A.Why not: Downstream models take upstream output as given; they do not validate it, so errors pass through.
Model interdependence and breadth of use raise materiality. A systematic upstream error is transmitted to every model and decision that depends on it, so it is a priority for validation.
In the app: 6 practice questions on this outcome, spaced review that brings it back before you forget it.
Start free →