The Credibility Report — Edition 34

June 12, 2026

AI-curated actuarial intelligence, designed by actuaries, for actuaries.

Opening Bell

This week's useful signal is capacity discipline rather than capacity scarcity. Everest is back in the catastrophe bond market for a large North America retrocession program, AM Best reports a strong Q1 underwriting rebound for U.S. P&C, and Gallagher Re's Southeast U.S. note points to cat bonds becoming ordinary reinsurance architecture. The research side is led by a directly actuarial LLM paper on extracting structured variables from unstructured claims documents.

This Week's Headlines

1. Everest targets $530m of North America retrocession through Kilimanjaro III Re

Everest Re is seeking at least $530m of multi-peril collateralized North America retrocession through two Kilimanjaro III Re 2026 catastrophe bond series. For actuaries, the read is not just "more cat-bond issuance"; it is that large reinsurers are using capital markets as a repeatable retro tool with multi-year risk-transfer economics.

2. U.S. P&C underwriting rebounds with a $16.3bn Q1 gain

AM Best reports that the U.S. property and casualty industry recorded a $16.3bn net underwriting gain in Q1 2026, compared with a $1bn loss in Q1 2025. This is a strong recovery signal, but pricing and reserving teams should still separate catastrophe normalization, prior-year development, line mix, and rate adequacy before changing selections.

3. Southeast U.S. catastrophe programs are leaning heavily into Rule 144A cat bonds

Gallagher Re's Schwebach highlights that Rule 144A cat bonds have reached roughly 80% adoption in Southeast U.S. property-cat programs, inverting the historical norm where traditional reinsurance carried most of the peak-peril load. That matters for ceded-cost allocation, attachment design, collateral mechanics, and basis-risk governance.

Research Spotlight

Paper of the Week: Leveraging LLMs for Unstructured Claims Data Analysis

The strongest actuarial research item this week is unusually direct: using LLMs to extract structured reserving, ratemaking, and claims-management variables from unstructured claim documents. The paper's two-stage pipeline separates document-level extraction from claim-level synthesis and validates core variables with clinical reviewers. The governance question is whether extraction accuracy, reviewer agreement, and downstream reserving impact can be monitored well enough for production claims analytics.

Time-series methods remain the common language across actuarial models

A new machine-learning review of time-series analysis is broad rather than insurance-specific, but it is relevant to mortality, lapse, inflation, claims frequency, and reserving work. Its value is as a technical map from classical state-space and ARIMA methods through Gaussian processes, ensembles, recurrent networks, and transformers.

Digital twins for sparse longitudinal disease progression

The Alzheimer's digital-twin paper is a health-methods watch item: sparse, irregular longitudinal observations with uncertainty-aware subject-level prediction are exactly the kind of setting life and health actuaries face in morbidity and claims pathways.

Backward coherence gives RNN hidden states a stability test

The RNN hidden-state paper proposes a quasi-reverse-martingale view of hidden-state stability. For actuarial sequence models, the practical question is whether this kind of stability regularization can make recurrent models more auditable when used for claims trajectories, mortality, lapse, or operational time series.

Tail-risk learning keeps moving toward large, dependence-aware models

ReSGA proposes a large model for Value-at-Risk and Expected Shortfall using retrieval-enhanced self-grouping autoencoders. It is finance-facing rather than insurance-facing, but the scaling and backtesting questions translate naturally into capital modelling, asset risk, and tail dependency governance.

Practical Takeaways

What We're Watching

Edition 34 • June 12, 2026