This is the framework for controlling a firm's models as a whole: it applies when a new model starts to be used, when the depth of validation is set, and when an audit checks whether the controls are effective.

Definitions of model and model risk

Model risk is the risk that a decision based on a model error, or on inappropriate use of a model, leads to an undesirable outcome. As examples of such outcomes, the Financial Services Agency (FSA)'s principles list a decline in the soundness of a financial institution's management, a violation of laws and regulations, and damage to the value of the firm1. The US SR 11-7 lists financial loss, poor business and strategic decisions, and damage to reputation2.

The scope of a "model" differs somewhat from one source to another. As the table below shows, the SR 11-7 definition includes "systems," and the 2026 revision excludes simple calculations. A mathematical method used for valuation or risk measurement counts as a model in line with the SR 11-7 definition.

Source Key points of the model definition
SR 11-7 (2011) A quantitative method, system, or approach that applies statistical, economic, financial, or mathematical theories, techniques, and assumptions to turn input data into quantitative estimates3
FSA principles (2021) A quantitative method that processes input data on the basis of theories and assumptions and produces estimates, forecasts, scores, classifications, and the like4
SR 26-2 revision (2026) A complex quantitative method, system, or approach that uses statistical, economic, or financial theories to turn input data into quantitative estimates5

The table cannot show everything, so the points below are added by source. A model under SR 11-7 has three parts: inputs, processing, and reporting. It is included even if the inputs are qualitative information or expert judgment, as long as the output is quantitative3. The FSA's principles also include methods in which some inputs or outputs are qualitative and methods in which the inputs rest on expert judgment4. The SR 26-2 revision excludes simple spreadsheet calculations and deterministic rule-based processing that is not built on statistical, economic, or financial theory5.

A model is a simplified representation of real relationships, and it rests on assumptions. There is no single correct method for a model, and the choice of method and assumptions can change the output a great deal6. A model that was appropriate when it first went into use can become outdated as the environment changes, and continuing to use it becomes inappropriate6. A firm that uses its own internal model under capital adequacy regulation has been required, as a condition of approval, to manage the risk of that model effectively7.

When model errors and misuse become risk

Model risk shows up mainly in two cases. In the first, judged against the use the model is meant for, the model's design has a fundamental error and its output is inaccurate. In the second, the model is used inappropriately18.

Errors can occur at any stage from design to implementation. Any of these can go wrong: the application of theory, the design of samples, the method of numerical calculation, the selection and estimation of inputs, and the implementation in information systems. Approximations and simplifications made to handle a complex problem can also reduce how far the output can be trusted8. If the input data or assumptions are wrong, the output is inaccurate even when the model's design is correct8.

Misuse can occur even with a model that has no errors. Risk rises when a model is used outside the environment it was designed for. This happens when an existing model is applied to a new product or market, and when market conditions or client behavior change and the model drifts outside its assumptions without anyone noticing9. The FSA's principles count use outside the intended range of use, and use beyond the model's limitations, as inappropriate use1.

Consider a hypothetical example in which a financial institution values transactions in its books at the model value. The model values one transaction at 1.0 billion yen. The table sets side by side a case in which the model error shows up as a valuation loss and a case in which it shows up as a valuation gain.

Case Model value Price actually concluded Difference
The actual price is below the model value 1.0 billion yen 0.95 billion yen Valuation loss of 50 million yen
The actual price is above the model value 1.0 billion yen 1.04 billion yen Valuation gain of 40 million yen

In both cases, the model fails to reproduce the price at which the transaction is concluded in the market. A valuation gain does not mean the model error was small. Users of a model may wrongly consider model risk to be low simply because the results of decisions based on the model appear favorable to the institution10.

Risk is also viewed in aggregate, not only model by model. When many models rely on common assumptions, data, or methods, a single problem can affect the output of several models at once11. The FSA's principles likewise call for addressing the interdependencies between different models12.

The three pillars of SR 11-7 and the revision

SR 11-7 is guidance on model risk management for banks, issued by the Board of Governors of the Federal Reserve System and the Office of the Comptroller of the Currency (OCC) on April 4, 201113. On April 17, 2026, the Federal Reserve Board, the FDIC, and the OCC issued a revised version, which replaced SR 11-714. The table below therefore follows the structure of the 2011 text, and the points that were revised come after it.

The text of SR 11-7 is organized around three sections (the pillars)15.

Pillar Key points of the text
Development, implementation, and use Development starts from a clear statement of purpose16
Validation Done by people who are not responsible for development or use and who have no stake in the model being judged valid17
Governance, policies, and controls The board of directors has ultimate responsibility and delegates execution to senior management18

The key points of each pillar are as follows.

  • Development, implementation, and use. The developer's experience and judgment influence how inputs and processing are chosen16. The opinions of model users are useful, but they lean toward the parts that bear directly on their own performance, and users find it hard to object to results that favor them10.
  • Validation. Work done by developers and users is also subject to critical review by an independent party. The elements are three: evaluation of conceptual soundness, ongoing monitoring, and outcomes analysis17.
  • Governance, policies, and controls. Policies are approved by the board or its delegates and reviewed every year. Internal audit assesses the effectiveness of the framework as a whole. A person is appointed to be responsible for the firm-wide model inventory18.

Model validation is the set of activities that check whether a model performs as expected in line with its design objectives and intended use19. SR 11-7 names "effective challenge" as the guiding principle of model risk management, including validation. It is critical analysis by objective, informed parties who can identify the model's limitations and assumptions and bring about appropriate changes. Whether the challenge is effective depends on a combination of incentives, competence, and influence20.

The revised version carries this framework forward and changes the following points.

  • Approach to operation. It sets out a risk-based approach scaled to the size of the model risk and to the size and complexity of the bank. The materiality of a model depends on how much its output affects business decisions (exposure) and on the model's purpose21.
  • Binding force. The guidance is not an enforceable standard, and failing to follow it is not by itself a ground for supervisory criticism. Supervisory action remains possible, however, where conduct violates laws or amounts to an unsound practice22.
  • Scope. It narrows the definition of a model (the table in the previous section) and places generative AI and agentic AI models outside the scope5.
  • Quality of validation. The quality of validation is determined by the rigor and effectiveness of the review, not by the organizational structure of the risk management function23.

The FSA's eight principles

On November 12, 2021, the FSA published its "Principles for Model Risk Management"24. They take a principles-based form instead of rules, and they assume that there is no uniform method of model risk management6. The FSA's stated policy is to put more weight on whether the framework as a whole works than on whether each item is observed. A firm can apply the principles flexibly according to the nature of its own risks and how it uses models and how complex they are25. The text of the principles is written with a focus on the three lines of defense of the three lines model. The first, second, and third lines in the table refer to these lines26.

Principle Content
1 Governance The board of directors and senior management build a framework that manages model risk comprehensively. They set policies and rules and define roles and responsibilities
2 Model identification, model inventory, and model risk rating Identify the models to be managed, record them in an inventory, and assign a risk rating to each model
3 Model development Set up a sound development process, prepare model documentation, and carry out model testing
4 Model approval Obtain internal approval when a model goes into use, when it is materially changed, and when it is revalidated
5 Ongoing monitoring After a model goes into use, check that it works as intended
6 Model validation Validators challenge the model effectively, independently of the first line, and carry out independent validation before use, at a material change, and after use begins (for validation by the first line, see the section on the three lines)
7 Vendor products and use of external resources Manage the risk when using vendor models or external resources
8 Internal audit The internal audit function assesses the effectiveness of the framework as a whole

The FSA says effective challenge is needed at every stage of the "model life cycle," which runs through identification, risk rating, development, use, change, and exit27. Principles 2 to 6 are arranged along this flow in the next section, a correspondence the FSA's principles do not state.

The flow from inventory to monitoring

As an example, the stages of the model life cycle in the FSA's principles run in the following order27. The steps are described by the line that performs them. The role of each line is covered in the next section.

  1. The financial institution identifies the models to manage, based on its own definition of a model. The first line usually does the identification, and the second line makes the final decision on whether something is a model28.
  2. Every identified model is recorded in the model inventory, each model is assessed for risk, and a risk rating is assigned. The assessment could take into account the model's materiality, complexity, and use29.
  3. In development, model documentation is prepared that describes the method, assumptions, and limitations. The first line carries out model testing before use begins30.
  4. The second line validates the model from an independent position (for validation by the first line, see the next section) and, taking the results of validation into account, the firm then obtains approval to use the model. The approver can attach conditions, such as limits on use, or refuse use31.
  5. After use begins, the first line performs ongoing monitoring and the second line revalidates. The second line can restrict or refuse use of a model in which defects are found32.
  6. When a material change is made to a model, additional validation is carried out as needed27.
First lineNot tied to a lineSecond lineIdentifymodelsMake the final callon whether it isa modelRecord ininventory, assignrisk ratingPreparedocumentationTest beforeuse beginsValidateindependentlyApprove useMonitor afteruse beginsRevalidate; limitor refuse useif defective
Figure 1Steps of the model life cycle and the three lines
Figure description

The figure arranges the steps of the model life cycle in lanes by the line that performs them. Steps 1 to 4 follow the arrows in order, and after use is approved the flow splits into monitoring by the first line and revalidation by the second line. "Not tied to a line" is the lane for steps where the text does not say who performs them. The first line usually does the identification of models, so it is placed in the first-line lane. Additional validation when a material change is made is not drawn, because it loops back to an earlier step; it is described in step 6 of the text. How validation is treated when the first line does it is described in the section on the three lines.

Consider a hypothetical example in which Bank A wants to apply an existing model, used for another product, to a new product. This is an example of use outside the model's intended environment9. The FSA's principles say that when a material change is made to a model, second-line validation and internal approval are obtained beforehand31. If applying the model to the new product counts as a material change, Bank A obtains validation and approval before using it, as in step 4 at the start of use. The additional validation at a material change corresponds to step 6.

The rating is the basis for deciding the depth and frequency of validation29. For models of low risk, the FSA's principles also allow revalidation on an irregular basis, for example when the environment changes significantly or there are signs of declining performance, instead of periodic revalidation33.

Consider how the FSA's principles apply to Bank A's models. For Model X, assessed as low risk, Bank A can choose not to revalidate periodically and to revalidate on an irregular basis when signs of declining performance appear. For Model Y, assessed as high risk, the frequency and depth of validation need to be commensurate with its risk33.

There are also exceptions in which a model is used before validation. The FSA's principles limit use without formal approval to a temporary measure under strict second-line controls34. The 2026 revision lists, for cases of urgent business need, paying attention to the model's limitations, informing the parties concerned, and deciding on controls such as limits on use and close watching of performance35.

The result of each step is kept in documents. The documents listed in the FSA's principles are as follows.

Document Content
Policies and rules Cover every item of model risk management, including the definition of a model, roles and responsibilities, the inventory, development, implementation, and validation36
Model documentation Describes the method, assumptions, limitations, and weaknesses to the extent that a third party, such as a model validator, can understand them30
Results of model testing The first line's results from checking how each component and the whole model behave and analyzing limitations and weaknesses30
Results of validation Documented and taken into account as criteria for the decision on model approval37
Records of ongoing monitoring The frequency, method, and results of monitoring32

Roles under the three lines model

As a framework for ensuring effective challenge in model risk management, the FSA's principles say that the roles of each unit could be organized under the "Three Lines of Defense (Three Lines Model)," which corresponds to the three lines model26. For the first and second lines, the roles from identification through monitoring and revalidation are described in the steps of the previous section. The table below concentrates on the roles that do not appear in those steps. The idea of the three lines model is covered in The Three Lines Model. Organizational structures and the assignment of roles differ from one financial institution to another, and in some cases separating the lines completely is not realistic26.

Party Role
Board of directors, senior management Build a framework that manages model risk comprehensively41
First line (model owners, model developers and users) Model owners are responsible for the use of the model and its performance2642
Second line (model risk management function) Maintains the model risk management framework and monitors model risk as a whole, from an independent position2642
Third line (internal audit function) Assesses the overall effectiveness of the model risk management framework from an independent position. Documents its findings and reports them to the board of directors or a related committee43

The board of directors may delegate implementation and maintenance of the framework to senior management or a related committee, and receives regular reports on compliance with policies and rules and on the state of model risk41. The second line also monitors compliance with policies and rules, assesses firm-wide model risk, and reports to the board of directors and others264227.

The FSA's principles call for a setup in which the people who validate models stand apart from the first line and can exercise sufficient challenge. Independence can be secured through separate reporting lines, the structure of incentives, and similar means. The first line may do the validation, and in that case the second line assesses the result44. Under SR 11-7, independence is not an end in itself but a means of aligning the purpose of validation with incentives. It can be supported by separate reporting lines, but is judged by actions and outcomes45.

The third line, internal audit, does not repeat the work of model risk management. It assesses whether management is comprehensive, rigorous, and effective. SR 11-7 and the FSA's principles agree on this point4643.

Scope of the US, Japanese, and UK guidance

The range of financial institutions covered differs by guidance. The SS1/23 of the UK's Prudential Regulation Authority (PRA) treats model risk as a risk in its own right47.

Source Issuer and date Scope Structure
SR 11-7 Federal Reserve Board and OCC, April 2011 Institutions for which the OCC is the primary supervisor, such as national banks, and institutions for which the Federal Reserve Board is the primary supervisor, such as bank holding companies and state member banks48 Development, implementation, and use; validation; governance, policies, and controls
SR 26-2 revision Federal Reserve Board, FDIC, and OCC, April 2026 Mainly banking organizations with more than USD 30 billion in total assets. Banking organizations below that size can also be affected where model risk is significant, judging from the number and complexity of their models and similar factors49 Development and use; validation and monitoring; governance and controls; vendor and other products
FSA principles FSA, November 2021 G-SIBs and D-SIBs in Japan, and the Japanese subsidiaries of G-SIBs that have obtained FSA approval for a model. The FSA may widen the scope50 Eight principles
PRA SS1/23 PRA, published in May 2023, applies from May 17, 202451 UK banks, building societies, and PRA-designated investment firms with internal model approval51 Five principles (model identification and risk classification; governance; model development, implementation, and use; independent model validation; and model risk mitigants)52

The SS1/23 definition of a model opens with the same sentence as SR 11-7 and also covers outputs that are qualitative53. SS1/23 places the method of dividing models into tiers according to risk in its first principle52.

What this page does not cover

This page does not cover how to build a single model or the valuation models themselves.

Notes

  1. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management) (November 12, 2021), "Ⅲ. 定義" (III. Definition) (b), page 5 (in Japanese). English translation published by the FSA, "Principles for Model Risk Management" (provisional translation, November 12, 2021): FSA. ↩ ↩2 ↩3

  2. Board of Governors of the Federal Reserve System and OCC, "Supervisory Guidance on Model Risk Management" (April 4, 2011, attachment to SR 11-7), "III. Overview of Model Risk Management" (page 3). ↩

  3. The same source, "III. Overview of Model Risk Management" (page 3), first paragraph. ↩ ↩2

  4. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management), "Ⅲ. 定義" (III. Definition) (a), page 5 (in Japanese). ↩ ↩2

  5. Board of Governors of the Federal Reserve System, FDIC, and OCC, "Revised Guidance on Model Risk Management" (attachment to SR 26-2, April 17, 2026), "II. Purpose and Scope" (page 3) and footnote 3. ↩ ↩2 ↩3

  6. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management), "Ⅰ. 意義" (I. Objective), pages 2–3 (in Japanese). ↩ ↩2 ↩3

  7. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management), "Ⅰ. 意義" (I. Objective), page 2 (in Japanese). ↩

  8. Board of Governors of the Federal Reserve System and OCC, "Supervisory Guidance on Model Risk Management", "III. Overview of Model Risk Management", the first of the two reasons (pages 3–4). ↩ ↩2 ↩3

  9. The same source, "III. Overview of Model Risk Management", the second of the two reasons (page 4). ↩ ↩2

  10. The same source, "IV. Model Development, Implementation, and Use", "Model Use" (page 7). ↩ ↩2

  11. The same source, "III. Overview of Model Risk Management" (page 4), "Aggregate model risk." ↩

  12. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management), "Ⅳ.（３）リスクベース・アプローチ" (IV. (3) Risk-based approach), page 7 (in Japanese). ↩

  13. Board of Governors of the Federal Reserve System and OCC, "Supervisory Guidance on Model Risk Management" (cover page, page 1). ↩

  14. Board of Governors of the Federal Reserve System, "SR 26-2: Revised Guidance on Model Risk Management" (April 17, 2026), the letter text and "Supersedes." ↩

  15. Board of Governors of the Federal Reserve System and OCC, "Supervisory Guidance on Model Risk Management", table of contents (sections IV, V, and VI). ↩

  16. The same source, "IV. Model Development, Implementation, and Use" (page 5). ↩ ↩2

  17. The same source, "V. Model Validation" (pages 9–11). Independence is on page 9, and the three elements are on page 11. ↩ ↩2

  18. The same source, "VI. Governance, Policies, and Controls" (pages 16–21). The board of directors and senior management are on page 17, policy approval and annual review are on page 17, roles are on page 18, internal audit is on page 19, and the inventory is on page 20. ↩ ↩2

  19. The same source, "V. Model Validation" (page 9), first paragraph. ↩

  20. The same source, "III. Overview of Model Risk Management" (page 4), "effective challenge." ↩

  21. Board of Governors of the Federal Reserve System, FDIC, and OCC, "Revised Guidance on Model Risk Management", "II. Purpose and Scope" (page 2) and "III. Overview of Model Risk and Model Risk Management" (pages 3–5). The relationship among materiality, exposure, and purpose is on page 4. ↩

  22. The same source, "I. Introduction" (page 2) and footnote 1. ↩

  23. The same source, "V. Model Validation and Monitoring" (page 8). ↩

  24. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management) (cover page and the last page, "令和３年 11 月 12 日 公表", published November 12, 2021) (in Japanese). ↩

  25. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management), "Ⅱ.（３）適用に当たっての留意事項" (II. (3) Points to note in application), page 4 (in Japanese). ↩

  26. The same source, "Ⅳ.（１）３つの防衛線" (IV. (1) The three lines of defense) (pages 5–6). ↩ ↩2 ↩3 ↩4 ↩5 ↩6

  27. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management), "Ⅳ.（２）モデル・ライフサイクル" (IV. (2) Model life cycle), pages 6–7 (in Japanese). ↩ ↩2 ↩3 ↩4

  28. The same source, Principle 2, 2.1 (page 9). ↩

  29. The same source, Principle 2, 2.2 and 2.3 (page 9). ↩ ↩2 ↩3

  30. The same source, Principle 3, 3.2 and 3.3 (page 10). ↩ ↩2 ↩3

  31. The same source, Principle 4, 4.1 (page 10), and Principle 6, 6.1 (page 11). ↩ ↩2

  32. The same source, Principle 5, 5.1 and 5.2 (page 11), and "Ⅳ.（２）モデル・ライフサイクル" (IV. (2) Model life cycle) (page 7). ↩ ↩2

  33. The same source, Principle 6, 6.5 (pages 12–13). ↩ ↩2

  34. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management), Principle 4, 4.2 (page 11) (in Japanese). ↩

  35. Board of Governors of the Federal Reserve System, FDIC, and OCC, "Revised Guidance on Model Risk Management", "V. Model Validation and Monitoring" (page 8). ↩

  36. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management), Principle 1, 1.3 (page 8) (in Japanese). ↩

  37. The same source, Principle 6, 6.1 (page 11). ↩

  38. Board of Governors of the Federal Reserve System and OCC, "Supervisory Guidance on Model Risk Management", "VI. Governance, Policies, and Controls", "Model Inventory" (page 20). ↩ ↩2

  39. The same source, "V. Model Validation", "Outcomes Analysis" (pages 13–14). Backtesting is on page 14. ↩

  40. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management), Principle 6, 6.3 (page 12) (in Japanese). ↩

  41. The same source, Principle 1, 1.1 (page 8). ↩ ↩2

  42. The same source, Principle 1, 1.4 (pages 8–9). ↩ ↩2 ↩3

  43. The same source, Principle 8, 8.1 (page 14). ↩ ↩2

  44. The same source, Principle 6, 6.4 (page 12). ↩

  45. Board of Governors of the Federal Reserve System and OCC, "Supervisory Guidance on Model Risk Management", "V. Model Validation" (page 9). ↩

  46. The same source, "VI. Governance, Policies, and Controls", "Internal Audit" (page 19). ↩

  47. Prudential Regulation Authority, "Model risk management principles for banks" (SS1/23, May 2023), 1.1. ↩

  48. Board of Governors of the Federal Reserve System and OCC, "Supervisory Guidance on Model Risk Management", footnote 1 (page 1). ↩

  49. Board of Governors of the Federal Reserve System, FDIC, and OCC, "Revised Guidance on Model Risk Management", "II. Purpose and Scope" (pages 2–3). ↩

  50. Financial Services Agency, "モデル・リスク管理に関する原則" (Principles for Model Risk Management), "Ⅱ.（１）対象金融機関" (II. (1) Scope of a financial institution), page 3 (in Japanese). ↩

  51. Prudential Regulation Authority, "Model risk management principles for banks" (SS1/23), cover page, 1.2, 1.3, and 1.5. ↩ ↩2

  52. The same source, the five principles in 3.1. ↩ ↩2

  53. The same source, Principle 1.1, (a). ↩