A practical approach to operational risk modelling: Scenario-based LDA

For financial services firms considering how to quantify Operational Risk (OR) capital under frameworks such as ICARA or ICAAP, one of the first questions is whether to adopt a modelling approach at all. We explore this broad question, including a discussion of the strengths and limitations of simpler approaches such as scenario summation, in our recent blog here.

For firms that have already decided to move towards a
modelling framework, a second question quickly arises: which modelling
approach should be used?

This is not a straightforward decision; Operational risk modelling often features limited data, reliance on expert judgement and high regulatory scrutiny. As a result, firms face a wide range of methodological options, from highly simplified techniques to complex statistical frameworks.

In this article we focus on one approach that is increasingly gaining traction across the industry: scenario-based Loss Distribution Analysis (LDA).

In our experience, scenario-based LDA approaches work well because they capture the core structural features that operational risk loss models must represent, while keeping the modelling framework as simple as possible in terms of implementation, explainability and validation.

Operational risk modelling should aim to capture the essential structural characteristics of operational risk while avoiding any additional statistical complexity that does not materially improve insight.

The table below summarises several structural features that any operational risk models must address.

These structural characteristics also highlight the key advantages of moving from simple scenario summation to a modelling framework. In particular, modelling allows firms to:

• Separate event frequency and loss severity, allowing different sources of data and expert judgement to be combined coherently.
• Extrapolate beyond historical experience and expert imagination, enabling plausible extreme losses to be represented through appropriate severity distributions.
• Capture dependencies between risk scenarios, allowing correlations and diversification effects to be incorporated consistently at portfolio level.

Scenario-based LDA provides a practical way to achieve these benefits while keeping the modelling framework relatively simple and transparent.

Some modelling frameworks attempt to introduce additional complexity beyond these structural features. In practice, however, doing so can increase model fragility without materially improving insight.

For many firms, the combination of analytical credibility, transparency and practical simplicity makes scenario-based LDA a compelling option.

Want to know more? Contact our team.

Operational Risk Modelling Can Be a Minefield

Operational risk modelling requires a combination of risk expertise, statistical capability and regulatory awareness. If these elements are not properly aligned, models can generate outputs that are difficult to interpret, validate or defend to regulators.

At the same time, firms face a wide range of modelling options — from highly simplified approaches through to complex statistical frameworks. Not all of these approaches are well suited to operational risk capital estimation.

For many organisations, the challenge is therefore not simply building a model, but choosing an approach that is robust, transparent and proportionate.

Why Scenario-Based LDA Works Well in Practice

Scenario-based LDA models operational risk losses using two core building blocks, which are elicited from experts in a structured scenario workshop process:

• Frequency — how many material loss events are expected to occur for each scenario in the next 12 months. When this is less than one, this is sometimes also expressed as the expected waiting time until the next material loss event (all else being equal);
• Severity — the potential financial impact when material events occur. This can conceptually be further split into:
  • Severity Magnitude — do losses tend to be in £1ks, £10ks, £100ks, etc – typically measured as a plausible high-severity loss e.g. worst event in 40 years; and
  • Severity Volatility — how much dispersion there is in the size of losses informing probabilities of exceptionally large events – typically driven by the ratio of the high-severity loss from above to a lower severity loss e.g. a median or typical loss.

Subject matter experts (SMEs) are typically asked to assess the types of events they could reasonably encounter during their professional careers, including both frequency and loss severity ranges. Statistical assumptions are then applied to extrapolate the full loss distribution, including the extreme outcomes relevant for capital estimation.

This structure allows firms to capture the wide range of loss outcomes that can arise from operational failures — from frequent small events to rare but severe losses.

Importantly, scenario-based LDA provides a practical way to capture these structural characteristics in a coherent and explainable way.

In practice, the statistical structure used within scenario-based LDA models is often deliberately kept simple. Many implementations rely on well-understood, industry-standard assumptions, such as Poisson distributions for event frequency, Lognormal distributions for loss severity and Gaussian copulas, rather than attempting to optimise every modelling choice. This reflects a conscious design principle: prioritising simplicity, transparency and explainability in pursuit of an estimate that is materially correct over unnecessary statistical complexity.

Making Data Useful — Without Letting It Dominate

One of the challenges in operational risk modelling is deciding how to use available loss data.

Internal and external loss datasets can provide valuable insight into operational risk. However, they reflect past experience rather than future risk, are often limited in size for low-frequency events, and may also be biased towards particularly good or bad fortune.

Scenario-based LDA offers a pragmatic way to incorporate this data without relying on it exclusively.

Loss data can be brought directly into scenario workshops to inform expert judgement, helping SMEs to anchor their assessments in real experience while still allowing them to consider how risks may evolve in the future. Practical approaches to incorporate data into scenario calibration will be discussed in more detail in our upcoming white paper on "Utilising data to enhance operational risk scenario and capital assessments for investment management firms."

More data-driven calibrations are also possible — for example, partially or fully calibrating model parameters to historical datasets. However, even in these situations firms must still address two important questions: how to ensure the model remains forward-looking, rather than simply extrapolating past losses, and how to calibrate risks where data is sparse or non-existent.

Combining data with structured expert judgement often provides a more balanced and credible outcome.

A Changing Regulatory Landscape

Operational risk capital continues to evolve within the regulatory framework. Key regulator developments shaping firms’ approaches include:

• PRA consultations on Pillar 2A capital frameworks (e.g. CP9/24 and CP12/25);
• updates to Basel 3.1 operational risk definitions and loss data expectations; and
• FCA observations following implementation of the Investment Firms Prudential Regime (IFPR).

Within this evolving regulatory environment, firms increasingly need approaches that are transparent, explainable and proportionate.

Scenario-based LDA can often meet these requirements while remaining adaptable to future regulatory developments.

Keeping Models As Simple As Possible

Operational risk teams already face a wide range of demands, including:

• assessing cyber and technology risks;
• monitoring operational risk appetites;
• upgrading operational resilience frameworks;
• evaluating control frameworks; and
• reporting risk exposures to boards and regulators.

Given these pressures, modelling approaches should aim to be as simple as possible.

Scenario-based LDA benefits from methodologies that are well-established across industry and feature a clear conceptual structure. This can significantly reduce model development timelines and help firms focus their effort on what matters most: understanding the organisation’s true risk profile.

Avoiding Common Modelling Pitfalls

When operational risk models become overly complex or poorly calibrated, three common problems often arise:

• Transparency – models are difficult to explain
• Plausibility – model outputs do not align with expert understanding of risk
• Stability – capital estimates change unpredictably

These challenges can undermine confidence in the model and create difficulties during validation or regulatory review.

By focusing on a modelling structure that captures the core drivers of operational risk — frequency, severity and scenario judgement — firms can often avoid many of these issues.

A Practical Path Forward

Scenario-based LDA offers a pragmatic way to move beyond scenario summation while keeping models understandable and defensible.

By combining expert insight, sensible statistical assumptions and thoughtful use of available data, firms can produce capital estimates that are credible, transparent and useful for decision-making.

In doing so, operational risk models become more than regulatory exercises — they become tools for understanding an organisation’s risk profile and prioritising investment in risk reduction.

For the same reasons discussed above, scenario-based LDA forms the foundation of Deloitte’s Capital Clarity operational risk modelling framework. The framework was developed to operationalise these principles in a practical and scalable way.

If you are interested in knowing more about anything you have read in this article or how we might support you in your operational risk data, scenarios and capital assessment journey, please reach out to the Capital Clarity team here or see our webpage: Capital Clarity | Deloitte UK.