Notes on risk, capital, and quantitative modelling.
Working papers, opinions, and explainers on IFRS 9, Solvency II, IRRBB, catastrophe risk, and the quantitative risk topics I spend my days thinking about.
AI Model Risk Management in 2026
AI adoption does not suspend model risk management. It raises the standard of evidence banks need before a model is allowed anywhere near a customer or a capital number.
4 min readPRA SS1/23 in Practice
SS1/23 is five principles long and deceptively demanding. The gap between a compliant framework and a working one is almost always ownership and evidence.
4 min readEU AI Act 2026: Transparency Duties
For financial institutions, the AI Act is less a new compliance regime than a documentation and classification problem attached to existing model governance.
4 min readGeopolitical Risk in ICAAP
Geopolitical risk is not a risk type. It is a shock generator that arrives through credit, market, operational and liquidity channels — and that is how ICAAP should model it.
3 min readIRRBB Beyond the Headline EVE Number
The EVE number is the output of a hundred assumptions, and two or three of them usually decide it. Volatile rates are when that stops being an academic point.
3 min readIFRS 9 Model Validation in Practice
Most IFRS 9 validation effort goes into re-checking the components. The risk concentrates in staging, scenario weights and the overlays nobody owns.
3 min readDORA: Lessons from Year One
The first year of DORA reporting revealed less about cyber attackers than about how badly most banks could describe their own systems under a deadline.
3 min readPrivate Credit and Bank Risk
Banks did not exit leveraged lending risk when private credit grew. They changed the form in which they hold it, and the new form is harder to see.
3 min readFrontier AI and Cyber Risk
Frontier AI does not invent new attack categories. It removes the cost and skill constraints that used to keep the existing ones rare.
3 min readEffective Challenge in the Second Line
Effective challenge is not a tone of voice or a meeting slot. It is the demonstrable ability to change a decision, and most frameworks cannot evidence it.
3 min readMachine Learning in Loss Reserving
Reserving did not lag pricing out of conservatism. The reasons are structural — and they tell you which new methods will work and which will not.
9 min readModel Risk & Independent Validation
Independent model validation is often reduced to a compliance exercise. That is a mistake — a well-run challenge function is one of the few controls that catches model failures before the market or the regulator does.
7 min readThe AI-Augmented Marine Actuary
AI will not replace the marine actuary. But marine actuaries who master AI — and the governance it demands — will replace those who do not.
3 min readCustomer Lifetime Value in Insurance Pricing
Lifetime value pricing carries more regulatory risk than the profession has acknowledged — but the naive version of the concern is wrong.
8 min readMarine Cargo Accumulation and AI Exposure
The largest marine losses of the modern era were accumulations nobody had measured. AI and maritime intelligence now make real-time exposure management possible — and soon, expected.
3 min readThe Connected Vessel and Hull Underwriting
For two centuries, hull underwriters priced ships on surveys, age and loss history — a snapshot taken once a year. Sensor-equipped vessels and machine learning are replacing the snapshot with a live feed.
3 min readGeographic Ratemaking Bias Under Climate Risk
As climate risk becomes better measured, territorial rating becomes simultaneously more accurate and more socially contested. That tension is structural.
8 min readCan a Language Model Reason Like an Actuary?
Not extract data. Not automate tasks. Support core actuarial reasoning. That is a much harder question than it sounds, and it is the right one.
8 min readMarine Technical Pricing with GLMs and ML
GLMs transformed motor and property pricing a generation ago. Marine is finally catching up — and machine learning is arriving at the same moment, offering a rare chance to leapfrog.
4 min readPredictive Modelling: Actuarial Meets AI
The pricing engine of a modern insurer is a hybrid of century-old actuarial technique and machine-learning tooling. Here is how the two fit together, and where the profession still gets the integration wrong.
7 min readActuarial Discipline in Marine Insurance
Marine is one of the oldest classes of insurance, yet one of the least actuarially penetrated. In a USD 40 billion market with thin margins and volatile losses, that has to change.
4 min readClaim Narratives Into Actuarial Variables
The headline is not that language models can read claim files. The contribution is the two-stage architecture — and the validation design most firms will skip.
9 min readCapital Modelling Under the Revised Solvency II Framework: Preparing for 2027
A practitioner tour of Solvency II capital modelling — standard formula, internal models, risk margin and the lambda factor — with a clear separation between the framework in force today and the revised framework that applies from 30 January 2027.
10 min readActuarial Communication for Risk Committees
The value of an actuarial analysis is realised when a committee makes a better decision because of it. That depends on how it is communicated at least as much as on how it is calculated.
4 min readModel Governance: Three Lines of Defence
Clear ownership, independent validation, and audit assurance make the difference between a model inventory that is trusted and one that is merely maintained.
3 min readClimate Risk Data Lineage and Proxies
Climate risk analytics are only as credible as the data behind them. Lineage, proxy hierarchies, and uncertainty controls determine whether the numbers can be relied on.
4 min readCredit Risk Model Monitoring Dashboard
Ongoing monitoring is where credit models earn their validation. A concise dashboard covering discrimination, calibration, stability, and overrides catches drift before it becomes a finding.
3 min readIFRS 17 Risk Adjustment Methods
The IFRS 17 risk adjustment is a disclosure with real consequences for CSM and profit emergence. Method choice matters; so does calibration discipline.
4 min readMarket Risk Stress Testing Beyond VaR
VaR summarises normal-market risk. Stress testing addresses what VaR cannot see. Both are needed, and neither is optional.
4 min readModel Risk Remediation and Finding Closure
A model finding is not closed when the owner says so. It is closed when the evidence supports the closure and the second line signs it off.
4 min readORSA Scenario Design and Board Use
An ORSA is only useful if its scenarios are severe, plausible, and specific enough to change a decision. Here is how to design them.
4 min readModern Loss Reserving: Chain-Ladder to ML
The chain-ladder is not obsolete and machine learning is not a silver bullet. This is a practitioner's map of how modern reserving actually combines the two, and where each fails.
6 min readEnterprise Risk Management in Practice
ERM frameworks look identical on paper. In practice the difference between one that runs the company and one that gathers dust comes down to a handful of design choices. Here are the ones that matter.
7 min readThe Geopolitical Premium in Supply Chains
Supply chain disruption is a material financial risk, not an operational headache. A framework for pricing the geopolitical premium and choosing mitigations.
5 min readIRRBB: EVE vs NII Sensitivity in Practice
EVE and NII look at the same balance sheet through very different lenses. Getting the mix right — and knowing when each one misleads — is what separates a real IRRBB framework from a compliance exercise.
3 min readSovereign Risk in a Fragmented World
Sovereign risk assessment now demands more than credit ratings. A four-pillar framework for finance professionals navigating a fractured geopolitical environment.
5 min readIFRS 9 PD Calibration: TTC vs PIT
IFRS 9 asks for a forward-looking, point-in-time PD. Regulatory capital asks for a stable, through-the-cycle one. Most banks have both, and reconciling them is where the real work lives.
4 min readLGD & EAD Validation: Common Pitfalls
PDs get the attention, but LGD and EAD errors quietly move ECL and RWA more than most banks realise. This is where independent validation actually earns its fee.
4 min readEnterprise Risk Management, Demystified
ERM isn't a binder on a shelf. It's the operating system a modern insurer or company uses to make decisions under uncertainty. Here's how to think about it.
6 min readGPD Tail Risk in Catastrophe Models
The tail is where catastrophe models earn their money and lose their credibility. Threshold choice, sample size, and parameter uncertainty matter more than the marginal-year fit almost anyone reports.
4 min readLiquidity Stress Testing Under LCR & NSFR
LCR and NSFR are the compliance floor. A useful liquidity framework is a set of plausible, severe, differentiated scenarios with pre-agreed actions — not a stack of green boxes.
4 min readLoss Reserving Explained
Loss reserving is one of the most consequential numbers on an insurer's balance sheet. This is a practical tour of how it's done and where the judgement calls hide.
6 min readPredictive Modeling in Insurance
Predictive modeling has quietly reshaped insurance pricing. Here's what actually works, what to watch out for, and how to keep models explainable to regulators.
6 min read