Model risk management faces a multitude of regulatory challenges, but it also needs to grapple with new complexities brought in by generative AI. How is generative AI making model risk management more difficult? Can generative AI be deployed by model risk managers for good? EXL shares their key findings from their two-part whitepaper.
In the UK, the PRA recently released its final policy PS6/23 and Supervisory Statement SS1/23 setting its expectations on MRM practices within banks. There are other regulatory guidance in the form of US Fed’s SR11-7 and ECB’s TRIM. While there are opportunities to learn and improve for banks through these regulations (within MRM), there is also a need to comprehensively understand regulations and define actions in simple, pragmatic, and practical terms.
To structure the ‘MRM challenge’, and specifically summarise key takeaways from regulations with pointed actions, EXL’s Risk Centre of Excellence has written a two-part white paper. In Part-1, we juxtapose prominent global guidance related to MRM and help define ‘minimum regulation’ and ‘industry-best’ practices. In Part-2, we define pointed actions on various items such as benchmarking, the role of the Board, the agility of MRM and managing business judgments within the MRM framework.
Model risk management has become increasingly complex in today's financial landscape, thanks to emerging technologies like generative AI. While these technologies offer exciting opportunities for innovation, they also present new challenges that banks and financial institutions must grapple with. This article will explore the challenges and opportunities in model risk management, drawing insights from EXL's white paper.
Challenges in model risk management
Coverage not commensurate to regulatory expectation
With the expanded use of deterministic quantitative methods (QMs) and data-driven decisions and inconsistently defined parameters, segregating models from non-models is becoming ambiguous.
Inadequate operating model and board oversight
Even if model development, validation and implementation are satisfactory, a weak governance function will significantly reduce the effectiveness of overall MRM. Board-level oversight is another major point PS6/23 dwells on. It suggests that the board should regularly receive reports on the bank’s model risk profile against its model risk appetite.
Improved MRM robustness yet poor execution speed
On one extreme, new lending firms place governance processes on the ‘back burner’, exposing them to compliance misuse and riskier business decisions. On the other hand, large banks with well-institutionalised processes and large teams are missing the need for speed to market to support business growth.
Generative AI complexity
Generative AI introduces complexity due to its ability to create synthetic data and mimic human-like responses. Banks must assess the risks associated with AI-generated data and the potential for biased or erroneous outputs.
Data privacy and ethics
Generative AI models can inadvertently generate sensitive or inappropriate content, raising concerns about data privacy and ethical use. Ensuring compliance with data protection regulations is paramount.
Gen AI opportunities in model risk management
- Enhanced efficiency: AI-powered tools can streamline model validation processes, reducing manual efforts and improving efficiency.
- Improved accuracy and robustness: Generative AI can enhance predictive modelling by generating more realistic and diverse scenarios, leading to more accurate and robust risk assessments.
- Innovation and competitive advantage: Banks can leverage generative AI to develop innovative model risk management tools, gaining a competitive edge in the market.
EXL’s six tips for enhancing model risk management
- Right-sizing MRM coverage and identifying model risks: Prioritise MRM for models and risk strategies by using complexity-materiality-impact-based risk tiers. Develop KPIs across the model’s lifecycle to better appreciate specific model risks.
- Addressing risks associated with gen AI models: Establish a comprehensive model risk management framework that encompasses all AI and generative AI models. Ensure it aligns with existing risk management processes.
- Regulatory compliance: Stay updated with evolving regulations related to MRM and the use of gen AI. Proactively adopt policies and procedures to remain compliant.
- Model performance and validation: Institute strong governance practices to monitor model performance including regular reviews and audits. Implement rigorous validation processes, especially for generative AI models, focusing on model performance and potential biases.
- Talent and training: Invest in training and hiring personnel with expertise in AI and generative AI, ensuring your team can effectively manage these models.
- Balancing organisational agility with acceptable MRM robustness: Revisit standards that cause delays in implementation or are sources of friction. Promote an environment of ‘robust actionable challenge’ based on expert opinion from risk assessment committees.
In conclusion
Generative AI presents both challenges and opportunities in model risk management for banks. While it introduces complexities and uncertainties, it also offers the potential for improved efficiency, accuracy, and innovation. By following the six tips from EXL and staying vigilant in adapting to evolving regulations, banks can navigate this dynamic landscape and harness the full potential of generative AI while effectively managing associated risks.
Learn more about Tips for Enhancing Model Risk Management at banks via this infographic.
