Model risk: what opportunities do models provide for financial services

For banks and insurers, models are extensively embedded in the business - from stress testing and liquidity forecasting to pricing, capital planning and investment strategy. As technology, business models, products and funding structures become more complex, reliance on quantitative models has accelerated. With this change comes a new reality: model risk is no longer a technical issue, it is a strategic one.

In our previous articles on liquidity risk and stress testing, we explored how speed, interconnection and execution under pressure now define resilience. Models sit at the centre of that agenda. They influence how risks are identified, how management actions are assessed under stress and how confidently boards can act in uncertain conditions. This article examines how model risk is evolving across Financial Services (FS) firms, the challenges firms continue to face and the opportunities created by strong more strategic model risk management.

Why model risk matters now

Models are essential for complex FS organisations and bring clear benefits. Automation, speed and analytical depth allow firms to assess risks that would be challenging to measure manually. Firms rely on models to simulate multiple variables under stochastic scenarios like market volatility, funding costs, behavioural responses, insured losses or collateral movements.

But as model use expands and reliance on models increases, so does the impact of model error, inaccuracy or invalidity. Poor data quality, judgement‑heavy assumptions, inappropriate modelling techniques or insufficient review and challenge can materially distort model outputs. Key strategic decisions are made from model outputs, and these failures can result in significant financial losses or reputational damage.

These failures typically arise from the following sources:

• Data: incomplete, stale or biased data can embed structural weaknesses into models long before stress emerges, particularly where proxies or limited histories are used.
• Model design: over‑simplified methodologies, inappropriate assumptions or models calibrated to benign conditions can break down when relationships between variables shift or become non‑linear.
• Implementation issues: can further distort outcomes, where coding errors, spreadsheet complexity or system limitations lead to outputs that diverge from the intended design.
• Governance and use risks: weak validation, insufficient independent challenge or overly optimistic overlays can reinforce false confidence in results. Models are also frequently applied beyond their original design purpose, with limitations not fully understood by decision‑makers.

Well known failures have shown how misplaced confidence in models can mask true exposures and accelerate losses once conditions turn. In the lead‑up to the financial crisis, the widespread use of models to price Collateralised Debt Obligations assumed stable and low default correlation; when housing markets turned, defaults became highly correlated, leading to a systemic underestimation of credit risk and sharp valuation losses across structured products. More recently, the collapse of Silicon Valley Bank highlighted weaknesses in interest rate and liquidity modelling, including assumptions around stable deposit behaviour, limited recognition of rapid rate rises on long‑duration assets, and an inability to capture the speed of a concentrated deposit run.

Managing model risk: common challenges across FS

Across the sector, several challenges persist. Models are often developed in silos, applied inconsistently across businesses or insufficiently prioritised by materiality. In some cases, firms rely heavily on complex outputs without clear accountability for assumptions, limitations or fitness for purpose.

Regulatory expectations have sharpened this focus. In the UK, SS1/23 (Model risk management principles for banks) has raised the bar for banks, expanding expectations around model definition, inventory management, tiering, validation and governance. While only formally applicable to banks using internal models, many FS firms now treat these principles as a baseline for good practice.

Firms making the most progress tend to:

• Apply consistent definitions - i.e. what constitutes a model and standards across all models.
• Prioritise models based on impact and complexity, with proportionate validation.
• Maintain strong second‑line and board‑level oversight, supported by meaningful model risk indicators.
• Place greater emphasis on data quality, recognising that even well‑designed models fail with poor inputs.

Crucially, leading firms treat model risk as a decision-support issue rather than a documentation exercise.

Banking and insurance: different pressures, shared lessons

Banks and insurers use models differently, reflecting distinct balance‑sheet dynamics. Banks operate in fast‑moving, liquidity‑sensitive environments, where models must adapt quickly to changing conditions. Insurers, by contrast, rely on models to understand how risks unfold over years, using them extensively for actuarial projections, asset‑liability management and ORSA stress testing. Insurers can leverage proven banking disciplines to strengthen model governance:

• Adopt clearer model inventories and tiering frameworks.
• Refresh assumptions and scenarios more frequently.
• Strengthen links between model outputs, escalation triggers and crisis execution.
• Improve operational readiness as balance sheets become more market‑sensitive.

What banks can learn from insurers

• Take a longer‑term perspective on resilience beyond short‑term survival metrics.
• Challenge judgement‑heavy assumptions more explicitly.
• Use richer, holistic scenario narratives to capture interdependencies.
• Apply multi‑year horizons to understand how vulnerabilities accumulate and inform capital planning.

Cross-sector learning reinforces the importance of aligning model outputs with how firms behave under stress.

Emerging challenges: AI and model complexity

The increasing use of Artificial Intelligence (AI) and machine‑learning models adds a further layer of complexity. Across FS, AI is now used in many areas, including pricing, underwriting, fraud detection and risk prediction. While powerful, these models raise new risks around explainability, data drift and embedded bias, particularly under stress, when historic patterns may no longer hold.

These developments reinforce a familiar lesson: strong governance, human oversight and effective challenge remain essential, regardless of model sophistication. Without them, AI can amplify existing weaknesses rather than strengthen resilience.

Model risk as a strategic opportunity

As with liquidity risk and stress testing, the firms that manage model risk most effectively are those that treat it as a core part of their resilience framework, not a compliance exercise. Models underpin key stress testing and liquidity calculations, including LCR, NSFR and recovery forecasts. Weak model governance in these areas can materially undermine a firm’s ability to prepare for, or respond to, stress.

Conversely, strong model risk management provides a strategic advantage. It gives boards confidence that decisions are based on robust analysis, that limitations are understood and that management can act decisively when assumptions fail.

In a financial services environment defined by speed, complexity and interconnected risks, model risk management is ultimately about confidence, confidence in data, in decisions and in the organisations ability to execute under pressure.