The Credibility Report
AI-curated actuarial intelligence — Edition 2
Welcome to Edition 2 of The Credibility Report — curated intelligence for actuaries who care about what's happening at the intersection of insurance, machine learning, and quantitative risk.
This week: 1/1 renewal deep-dive, negative binomial reserving models, spatiotemporal health risk assessment, PRA streamlining, and our research spotlight on interpretable ML.
📊 The 1/1 Renewals: A Softer Landing
The January renewals are in, and the picture is nuanced: retrocession renewed "very late" with capacity pressures affecting rates, but overall reinsurer profitability remains robust—if lower than 2024's exceptional results.
🔥 Retrocession capacity tightened
Renewals came down to the wire with capacity constraints pushing rates higher on frequency-affected accounts. ILS inflows and retained earnings provided critical support, but the market is clearly more selective.
🚗 Sidecars in focus
GC Securities executives are flagging sidecars as a major theme for 2026, with aggregate reinsurance seeing renewed interest as cedants seek more flexible structures.
📅 Multiyear deals return
In a notable shift, reinsurers are responding to cat bond competition by offering multiyear terms again, particularly for clean accounts with strong loss experience.
By the Numbers
| Metric | Value | Source |
|---|---|---|
| US PRT premiums (2025 est.) | Up to $50bn | Aon |
| Matterhorn mortality bond | $46.9m early return | Swiss Re |
| P&C Q3'25 underwriting | Improved results | Industry |
The ILS Angle
Swiss Re has been active on both sides of the mortality/property divide:
- • Matterhorn mortality cat bond redeemed early, returning $46.9m to investors
- • New $125m peak NA peril aggregate retro sought through Matterhorn Re
Meanwhile, private ILS growth continues alongside cat bonds and ILWs. Culpeper Capital is eyeing casualty ILS expansion, seeing the sector "ripe for growth."
"1/1 renewals signal lower, but still robust reinsurer profitability for 2026." — Artemis.bm
📚 Research Spotlight
Negative Binomial Models for Development Triangles of Counts
arXiv | Read the paper →
A return to fundamentals with modern rigor. This paper develops negative binomial models for claims count triangles, extending the classic chain ladder framework with proper handling of overdispersion and Bayesian inference.
Why actuaries should care
Claims count modeling often gets less attention than severity, but it's fundamental to reserving accuracy. This paper bridges classical actuarial methods with modern Bayesian approaches.
Practical applications
IBNR count estimation, reserve variability, integration with existing chain ladder workflows.
Other Notable Papers
Spatiotemporal Actuarial Risk Assessment of Health Portfolios Using Copula-Based Bayesian Neural Networks
ResearchGate | Read →
Combines copulas, Bayesian neural networks, and spatiotemporal modeling for long-term care insurance solvency assessment.
💡 Key insight: Environmental and climate factors increasingly matter for health and LTC portfolios—not just property lines.
Explaining Machine Learning Predictive Models through Conditional Expectation Methods
arXiv | Read →
A deep dive into conditional expectation-based explainability methods for ML models. Relevant for actuaries navigating interpretability requirements.
💡 Why it matters: Regulators increasingly require model explanations. This provides theoretical grounding for SHAP-like approaches.
Mixtures of Transparent Local Models
arXiv | Read →
Proposes mixtures of interpretable local models as an alternative to black-box ML. Maintains transparency while capturing complex patterns.
💡 Practical applications: Pricing models where full interpretability is required, regulatory submissions, model validation.
The Limits of Complexity: Why Feature Engineering Beats Deep Learning in Investor Flow Prediction
arXiv | Read →
A sobering reminder that deep learning isn't always the answer. In this domain, careful feature engineering outperforms more complex approaches.
💡 Why actuaries should care: Reinforces that domain expertise + simple models often beats throwing neural networks at problems. Sound familiar?
📄 From the arXiv
Top-scored preprints from the past week:
| Paper | Score | Keywords | Link |
|---|---|---|---|
| Tab-TRM: Tiny Recursive Model for Insurance Pricing | 25 | insurance pricing, tabular, ML | → |
| Penalized Likelihood for Adaptive Neighborhood Clustering | 27 | mortality, hazard, interpretable | → |
| Time Series Foundation Models for Agricultural Forecasting | 22 | deep learning, forecasting | → |
| Dual-Level Physics-Informed Time Series Forecasting | 22 | neural network, LSTM | → |
| DeePM: Regime-Robust Deep Learning for Portfolios | 22 | transformer, risk | → |
| RL of LLMs for Interpretable Risk Assessment | 21 | reinforcement learning, interpretable | → |
🏛️ Regulatory & Professional Updates
PRA 2026 Priorities: Streamlining Ahead
The Bank of England's PRA has announced plans to streamline supervision as part of its 2026 priorities. Key themes include reducing regulatory burden while maintaining robust oversight.
Read the announcement →EIOPA Strategy 2030
EIOPA has unveiled its new strategic direction through 2030, focusing on sustainable finance integration, digital transformation, and proportionate supervision.
View the strategy →UK-EU Regulatory Cooperation
Post-Brexit cooperation continues: UK and EU regulators have signed a new Memorandum of Understanding to facilitate cross-border supervision and information sharing.
Read more →📅 Conference Calendar
| Event | Dates | Location |
|---|---|---|
| ASTIN Colloquium 2026 | TBC | TBC |
| CAS Spring Meeting | May 2026 | TBC |
| Monte Carlo Rendez-Vous | September 2026 | Monaco 🇲🇨 |
| Baden-Baden | October 2026 | Germany 🇩🇪 |
🔮 What We're Watching
Q1 cat activity
Will the benign 2025 trend continue?
Casualty ILS expansion
Culpeper and others pushing into new territory
IFRS 17 Year 2
First full annual results and lessons learned
AI regulation evolution
State-level developments in the US