What Is Model Risk Management? (With Types and Tips)

By Indeed Editorial Team

Published January 3, 2022

The Indeed Editorial Team comprises a diverse and talented team of writers, researchers and subject matter experts equipped with Indeed's data and insights to deliver useful tips to help guide your career journey.

A growing reliance on models, regulatory challenges, and talent scarcity is encouraging banks to take a more value-centric risk management strategy. Model risk refers to situations when a financial model used to measure a firm's market risks performs inadequately. Implementing an effective model risk management (MRM) system is critical for organizations that rely on quantitative operations and decision-making models. In this article, we answer "what is model risk management," offer tips to improve your strategy, discuss profit and loss, list different types and causes of model risk, and discuss why it occurs.

What is model risk management?

To help answer the question "what is model risk management?", it's beneficial to review the definition. A model is a quantitative and mathematical system used to convert input data into quantitative estimates, and model risk management is the monitoring of risks associated with using models incorrectly. Model risk management involves using techniques, practices, or behaviours to identify and measure model risks so that a company or organization can avoid them.

Financial institutions, such as banks and insurance companies, often rely on credit, market, and behavioural models for various purposes that involve almost all of their daily operations. These models have developed into vital risk management and operational efficiency strategies. Financial institutions make money primarily by taking risks, so they can benefit significantly from using more effective models to analyze risks, understand customer behaviour, define funding requirements, assess capital adequacy, make investment decisions, and manage data analytics.

Tips for strengthening your model risk management approach

Financial institutions that employ predictive, financial, or economic models may consider strengthening their MRM approach. Here's a list of tips to help you engage in MRM:

  • Compile an inventory of existing models. It's essential to make an inventory of any current or in-development models to track each and its uses. You can clarify the difference between a model and a tool so that all stakeholders understand how to use and contribute to the inventory, including the purpose, model owner, data sources, and significant assumptions for each model.

  • Understand regulatory requirements related to model use and verification. This understanding can assist the business in managing entity-wide risks and establishing MRM processes that comply with legislation that may affect them in the future.

  • Use model testing and validation. Institutions can test and validate any substantial or complex models before installation so that management can be confident in model outputs. You can check the model regularly to identify potential errors, track corrective measures, and ensure acceptable use based on the outcomes of the testing process, which individuals or an independent third can supervise and validate.

  • Involve stakeholders. The board can oversee the entire MRM process, while management can establish the MRM framework and related procedures. Leaders with organizational knowledge can take part in the MRM process to ensure that assumptions remain reasonable, model documentation is accurate, and data sources are legitimate.

Related: What Is Unskilled Labour? (With Examples and Salaries)

Profit and loss

In terms of profit and loss improvement, MRM can decrease rising modelling costs by addressing fragmented model ownership and processes caused by complicated models. This can often save financial institutions a lot of money. Banks can better align model investments with business risks and priorities by improving their understanding of the model landscape. MRM can reduce some profit and loss instability by lowering model risk and minimizing its impact. Overall, the effect can increase model transparency, allowing you to redirect cost-cutting resources to higher priorities.

Systematic cost reduction only works with an end-to-end MRM approach. This method aims to streamline and automate essential modelling operations, saving model-related expenditures by 20% to 30%. For example, banks are increasingly attempting to manage the model-validation budget, which has risen due to increased model inventories, improved quality and consistency requirements, and higher talent expenses. The industrialization of validation processes, which employ lean principles and an efficient model-validation technique, has these options:

  • Prioritization with 30% savings: Validation models get selected based on factors such as their significance to business choices. To improve speed and efficiency, model tiers can change the strength of validation or help you create new resource strategies and governance methods.

  • Portfolio management office and related tools with 25% savings: With standardized methods, tools, and governance mechanisms, you can eliminate inefficiency at each stage of the validation process. These may contain standards for development, submission, validation strategies, and playbooks.

  • Coding and testing with 25% savings: Automating well-defined and recurring validation processes, such as standardized testing or model replication, can further reduce costs.

Related: How to Write an Action Plan to Help You Achieve Your Goals

Types of model risks

Here are three types of model risks:

Model specification risk

Model specification risk occurs primarily during the design process. Minor errors can result in a specific combination of inputs and computations that produce misleading or otherwise unusable outputs. While an experienced modeller may review a design document and speculate on how well the model might perform, they may complete the process by empirically evaluating its performance.

Model specification errors can also occur after a developer has finished designing a model. The model itself may be modified from time to time, to address new traded instruments, correct a perceived problem with its performance, or stay current with industry practices. It can also occur with no noticeable changes to a value-at-risk measure. For example, you can get historical values for a specific key factor from a time series kept by a data vendor. If the data vendor changes how it calculates the importance of that time series, it may affect the performance of the value-at-risk measure.

Related: Smart Goals: Objectives for Your Career

Model implementation risk

Model implementation risk is the risk that a model may deviate from what the developers defined in the design document once they implement it. This is because inputs may vary. For example, you can get historical data for a critical factor from a source other than the one specified in the design document. Mathematical formulas can differ because of simple typing errors or because a programmer implements a formula in a way that inadvertently changes it. The risk also is due to the chance that outputs are misrepresentative. For example, two numbers juxtaposed in a risk report may seem confusing.

If the implemented model deviates from its design, regardless of how well it functions, it differs from what the creator and users understand. They believe they have one model when, in fact, they have two. This means the outcome may be uncertain. Human error is the most common cause of implementation error. Some code or logic errors are almost unavoidable in large software projects. Implementation risk also develops if dishonest programmers choose to take shortcuts. Unfortunately, such dishonesty occurs, although it's uncommon in large-value-at-risk solutions that undergo rigorous testing and validation.

Model application risk

Model application risk is the possibility that a model may get misunderstood or misused. Misinterpretation can create a misleading sense of security or prompt less advisable actions. Modellers typically come from different backgrounds than traders or senior executives. While model designers can understand the meaning of their outputs on a technical level, the model's users may have a more intuitive, less precise understanding. Users may have a better sense of how reliable a model's outputs are than modellers who may be less familiar with the specific information being modelled.

If they disagree with what a model shows, users may be more likely to decide based on their emotions than on the model's advice. Another common type of model application risk is pairing an otherwise effective value-at-risk measure with a portfolio that's not appropriate for that measure. This can occur if an organization's trading activities grow over time, but they don't update the value-at-risk assessment to reflect new instruments or trading techniques.

Related: Tips for Training New Employees Effectively (With FAQ)

What causes model risk?

Model risk typically happens either because the model is unfit for the purpose or because it contains theoretical and implementation flaws. Developers are typically responsible for thorough documentation during model creation, which they can regularly update as the model and its application environment change. Model users and developers who understand the models fully may be hesitant to document because of the work and length of time required. Banks can give incentives for developers to generate full model documentation to lessen the possibility of a model risk occurring.

Banks can also ensure that other participants in model risk management activities, such as benchmarking, outcomes analysis, continuous monitoring, and process verification, document their work. The line of business or other decision-makers can document the information regarding the selection of a specific model and the validation that follows. For example, when banks use models from vendors, they can ensure that appropriate documentation of the third-party approach is accessible to determine the model's proper validation.

Explore more articles