Cairn Underwriting is a managing general agent based in Edinburgh, writing specialty property and marine risk across the UK and Northern Europe. Their senior underwriter, David, has been writing Norwegian offshore energy risks for fourteen years. He knows the North Sea platforms, the operators, the loss history, and the seasonal risk patterns better than any model does.
A new submission arrives: a large production platform with USD 180M TIV. The cat model run by Cairn's junior analyst, Anya returns a 1-in-100 year OEP of USD 31M and an AAL of USD 890K. David looks at the numbers and shakes his head. "The model is underestimating this. That platform is old, poorly maintained, and in a zone where the model's storm data is thin. I'd put the real number at least 40% higher."
Anya is uncertain. But David has fourteen years of experience telling him something different. Who is right and more importantly, whose view should determine the price?
Every cat model output is a starting point. It tells the analyst and the underwriter, what a probabilistic model calculated given a set of inputs, assumptions, and physical science calibrations. What it does not tell anyone is whether to accept the risk, at what price, in what structure, or with what conditions. Those are decisions. Decisions require judgement. The model provides essential input to that judgement but cannot replace it.
This is not a limitation of current cat models, it is a structural feature of what models are. A cat model is a quantitative framework built on historical data, physical science, and calibration assumptions. It knows what it has been calibrated to. It does not know the maintenance history of a specific offshore platform. It does not know that a new building's contractor cut corners on the roof connections. It does not know that a city's drainage system has deteriorated since the flood maps it is using were drawn. The practitioner carries that knowledge.
The model is one input into a professional view of risk. It is typically the most important quantitative input the anchor for the expected loss calculation and the tail risk estimate. But it sits alongside market knowledge, exposure data quality assessment, non-modelled item evaluation, and underwriting experience. A decision made purely from the model ignores what the model cannot know. A decision made purely from experience without the model ignores the discipline the model provides. Professional cat modelling practice is the combination of both.
Between the moment Anya runs the model and the moment David decides whether to bind the Norwegian platform, several things happen. Each step adds information or context that the model alone could not provide. Understanding this journey is what allows a new practitioner to understand where their work fits in the larger professional process.
Anya runs the stochastic catalogue against the platform's exposure data. The model returns an AAL of USD 890K and a 1-in-100 year OEP of USD 31M. These figures reflect the physical science of North Sea storm hazard, the platform's TIV and reported construction class, and the financial terms of the policy. The model has done everything it was designed to do.
Anya reviews what the model was given. The construction class was reported as "offshore steel platform" a broad category. The year built is 1987 older than many of the damage functions were calibrated to. The geocoding is accurate to platform coordinates. She notes that the model's North Sea storm catalogue is thinner than its UK onshore catalogue. She documents these observations. The model number is reliable within the limits of what it was given.
David brings years of North Sea experience to the number. He knows the operator's maintenance record from previous submissions. He knows the platform's structural inspection history. He knows that the zone it operates in has seen more severe storm intensity in recent years than the historical average the model's catalogue was built on. None of this is in the model. All of it is relevant to the actual risk.
David and Anya together form a view: the model's USD 31M 1-in-100 year figure is a reasonable quantitative anchor, but given the data quality gaps and the platform-specific concerns, the practical view of risk warrants a loading. They agree on a view of approximately USD 43M at the 1-in-100 year level, a 39% upward adjustment which feeds the pricing recommendation. The decision to bind, decline, or seek more information rests with David. The model informed the starting point. The professional judgement determined the view.
In professional cat modelling practice, a "view of risk" is the considered position on a risk's expected loss and tail exposure that an organisation adopts for pricing and accumulation management. It is not synonymous with the model output. It incorporates the model output as the primary quantitative input and then layers on the additional context data quality assessment, non-modelled item evaluation, market knowledge, and underwriting experience that the model alone cannot provide.
1-in-100 year OEP: USD 31M based on stochastic storm catalogue
AAL: USD 890K from probability-weighted event losses
Physical damage only no BI, no pollution liability
Generic offshore steel construction damage functions
Platform age and maintenance record suggest higher-than-default vulnerability
Operator's historical loss frequency above market average known from prior submissions
Recent storm intensity trend in this zone
Non-modelled BI and pollution liability exposure needs separate assessment
The view of risk is what the underwriter takes into the pricing conversation. It may be numerically close to the model output if the data is good, the peril is well-calibrated, and no significant contextual factors pull in either direction. It may diverge substantially if there are data quality gaps, non-modelled exposures, or market intelligence that changes the picture. What matters is that the view is explicit and reasoned not a gut feeling overriding the model, and not blind acceptance of the model ignoring everything else.
There are legitimate reasons to form a view above or below the model output. Understanding when those reasons apply and when a divergence from the model is instead a red flag is one of the core professional skills the cat modelling practice develops over time.
Data quality gaps in the exposure: incomplete COPE, approximate geocoding, or missing TIV components mean the model is working with less information than the risk actually presents. The model's output may understate the risk simply because it did not receive accurate inputs.
Known non-modelled items: significant BI exposure, demand surge risk in a concentrated geography, secondary perils the model does not capture, or litigation risk in a market known for loss amplification all justify a view above the raw physical damage output.
Market intelligence the model cannot access: knowledge of a specific operator's poor maintenance record, a building's known structural deficiencies, or a location's local drainage problems that are not reflected in the model's hazard grid.
Confirmed risk mitigation the model defaults conservatively on: a newly retrofitted building the model treats as older stock, a sprinkler-protected warehouse the model applies a standard fire loading to, or a coastal property with documented flood barriers the model's grid does not capture.
Model conservatism in thin-data regions: for some geographies and perils, models apply conservative defaults where data is sparse. If an independent assessment of the risk produces a more optimistic view, the professional judgement may sit below the raw model output.
Good data, well-calibrated peril, no material non-modelled items: a well-geocoded modern building in a mature insurance market with confirmed COPE and accurate TIV, assessed on a well-calibrated peril like US hurricane windstorm. Here the model output is often the most reliable single estimate available and diverging from it requires explicit justification.
Adjusting the view of risk above or below the model is not something that should be done casually or without documentation. A professional view of risk states the model output, identifies the reasons for any adjustment, quantifies the adjustment where possible, and records the reasoning. An adjustment that cannot be explained and documented is not professional judgement it is overriding the model for reasons the organisation cannot assess or audit
In most professional cat modelling environments, model outputs do not go directly to the underwriter. They pass through a review and sign-off process, a check that the run was configured correctly, the data quality is understood, the limitations have been identified, and the output is fit for the purpose it is being used for. This process exists because the downstream consequences of a model run, pricing decisions, capital allocation, regulatory submissions are consequential enough for verification.
Sign-off is not a guarantee that the model is right. It is a confirmation that the model was used correctly, with known-quality inputs, and that the limitations of the output have been appropriately characterised. A signed-off model run on poor-quality exposure data is still a model run on poor-quality data. Sign-off documents does not correct what was missing.
Understanding what sign-off means and what it does not mean, matters for practitioners at every level. A junior analyst preparing a run for sign-off needs to understand that their responsibility is not just to produce a number but to document what the number does and does not reflect. A senior analyst reviewing a run for sign-off is confirming that the work was done correctly and that the caveats are complete. Neither is certifying that the output is the truth, only that it is what the model produced given the inputs it received.
The model produces an output. The decision is made by the underwriter, who uses the output as one input among several. A decision made purely from model output ignores everything the model cannot know. No professional pricing or risk acceptance decision is or should be made by handing the model output directly to the binding authority without contextualisation.
Experience provides irreplaceable context. It does not provide a probability distribution over thousands of synthetic events calibrated to the physical science of hazard. David's fourteen years of North Sea experience tells him the model is underestimating, it does not give him an AAL or an EP curve. The model and the experience are complements, not substitutes.
Adjusting without documented reasoning is unprofessional. Adjusting with explicit, reasoned, documented justification is exactly what the professional view of risk is supposed to represent. The model is calibrated to general patterns; the professional view incorporates specific knowledge the model cannot access. Both are necessary.
Sign-off means the output was correctly produced from the inputs provided, and the limitations are documented. It does not mean the inputs were accurate, the model was the right tool for this risk, or the output will match the actual loss. Sign-off is a process discipline, not an accuracy guarantee.
The considered professional position on a risk's expected loss and tail exposure, incorporating model output plus additional context on data quality assessment, non-modelled item evaluation, and underwriting experience. The basis on which pricing and acceptance decisions are made.
The numerical result produced by a cat model run AAL, EP curve, return period losses. The primary quantitative input to the view of risk.
The professional assessment brought by an experienced underwriter to supplement model output, knowledge of specific risks, operators, markets, and loss patterns that no model can access. Works alongside model output to form the view of risk.
An upward adjustment applied to a model output to account for known limitations, non-modelled items, data quality gaps, or specific risk factors that justify a view above the base model estimate. Must be documented and reasoned to be professionally defensible.
The review and approval process confirming that a model run was correctly configured, the data quality is understood, and the limitations are documented. Confirms the work was done correctly.
The authority to formally accept a risk and issue a policy. In most underwriting environments, the cat modelling analyst does not hold binding authority they inform the underwriter's decision, which is exercised by the person with delegated authority to bind.
Five questions · 4 of 5 correct (80%) to pass