Managing Model Risk in Electronic Trading Algorithms

Trading algorithms are based on quantitative models, which present risks that need governance. Regulators have provided general guidance on model risk management, but applying this to algorithmic trading requires a tailored approach.

To address this need, the Financial Markets Standards Board (FMSB) recently released a draft “Statement of Good Practice” (SoGP) to support firms proportionately applying model risk management frameworks to models deployed in their electronic trading algorithms (‘Algos’). The SoGP considers the nature, scale and complexity of such models as well as existing systems and risk controls intended to mitigate associated market, conduct, credit and operational risks.

This blog summarises the crucial considerations for firms in managing the model risk associated with electronic trading algorithms detailed in the FMSB SoGP.

The FMSB, in its SoGP, outlines five key areas mapped to nine Good Practice Statements (GPS) on implementing model risk management to algorithms proportionately used in electronic trading.

The five key areas covered in the SoGP are:

FMSB proposes nine Good Practice Statements on implementing model risk management to algorithms.

The nine Good Practice Statements are summarised below:

1. Identifying Models in Algorithms

Trading algorithms need to be examined to identify any components that meet the criteria for a "model". FMSB defines this as the method applied to produce a quantitative estimate; and applies statistical, economic, financial or mathematical theories, techniques and assumptions and does not consider simple calculations to meet the definition of models.

2. Categorising Model Risk Tiers

Not all models pose equal risks, and higher-risk models require more scrutiny. Models should be assigned risk tiers based on factors such as uncertainty around model outputs, complexity, criticality of the model and speed of performance feedback.

3. Tailoring Model Testing

Testing should assess how models perform under volatile conditions and with limited data. The focus should be on testing controls that mitigate model inaccuracies.

4. Considering Controls When Validating Models

The controls implemented around algorithms, like trade limits, help reduce model risk. Validation should factor in these controls and their risk mitigation rather than solely focusing on model accuracy. It should include:

• Embedded controls in the model logic
• Wider controls like trade limits outside the model
• Manual trading supervision
• Monitoring algorithmic trading metrics

5. Align Validation to Model Risk Profile

Higher-risk models require more intensive validation. But, for algorithms, assessing actual model behaviour is better than peer comparison as the model undergoes constant changes.

6. Ensuring Knowledgeable Validation Staff

Staff validating algorithmic models need market expertise. Firms may need to define roles and responsibilities to achieve the expected results.

7. Monitoring Aligned to Model Risk Tier

Higher-risk models require more frequent performance monitoring. However, firms need to use judgement in responding to monitoring alerts and consider material adverse outcomes.

8. Expediting Validations for Needed Model Changes

When models need updates to address changing market conditions, firms can defer validation until after implementation if they review the change impact and have controls in place.

9. Leveraging Source Code Access for Documentation

Frequent model changes make comprehensive documentation difficult. Providing access to model source code can supplement documentation and enable independent validation when code changes frequently.

Conclusion and Next Steps

The FMSB guidance aims to balance innovation in algorithmic trading with due consideration of model risk principles. Below is a summary of next steps for firms wishing to identify gaps against FMSB guidance.

Should you wish to discuss this topic further or require support with considering the risks posed by Algorithms and AI and the necessary enhancements to your Control Framework, please don’t hesitate to get in touch with our AI & Algorithm Assurance team here.