The Credibility Report
Edition 29: Soft Markets, Alternative Capital, and Algorithmic Pricing Fairness
Aon points to buyer-friendly but segmented market conditions, Guy Carpenter frames alternative capital as reinsurance architecture, Triple-I and NCCI sharpen the P/C profitability picture, Aviva reports GI growth with ratio improvement, and arXiv surfaces a directly relevant fairness-testing paper for deterministic pricing algorithms.
This week’s insurance-market signal is segmentation: softening is real, but only where risk quality, attachments, and evidence support it.
The research signal is sharper than usual for actuaries: deterministic pricing algorithms need fairness tests with valid inference, not comforting audit theatre.
Opening Bell
This week’s signal is not simply “more softening”. It is segmentation: property and reinsurance markets are becoming more buyer-friendly where the data, attachments, and risk quality are credible, while casualty, cyber, geopolitical, and climate exposures are still forcing more granular underwriting evidence. On the research side, the strongest arXiv item is unusually close to actuarial practice: algorithmic pricing fairness tests that argue the standard audit regression is statistically invalid for deterministic pricing systems.
Key Metrics
| Metric | Latest signal | Direction | Why actuaries should care |
|---|---|---|---|
| Global commercial insurance | Aon says Q1 2026 stayed broadly buyer-friendly, with available capacity and competitive pricing | Softer, but segmented | Pricing indications should separate market movement from risk-quality selection effects. |
| Alternative reinsurance capital | Guy Carpenter frames cat bonds, sidecars, and financial investors as redesigning reinsurance capital architecture | More structural convergence | Capital-cost assumptions increasingly depend on model transparency, attachment clarity, and collateral mechanics. |
| U.S. P/C market | Triple-I/Milliman report the lowest net combined ratio in more than a decade in 2025 | Underwriting recovery | Planning should not extrapolate the headline: commercial auto and GL remain above 100 NCR. |
| Workers’ compensation | NCCI reports a 2025 calendar-year combined ratio of 91 and accident-year combined ratio of 102 | Prior-year releases still matter | Reserve strength is still doing work; accident-year adequacy needs separate scrutiny. |
| Aviva GI | Q1 2026 general-insurance premiums up 19% to £3.4bn; undiscounted combined ratio improved to 94.1% | Growth with margin repair | Peer benchmarking should normalize for mix, acquisition effects, and rate-cycle timing. |
| arXiv scoring | 14 actuarial-relevant papers scored after increasing the arXiv API timeout; top score 42 | Research pipeline refreshed | The pricing-fairness audit paper is directly relevant to model governance and regulatory evidence. |
Headlines
1. Aon: buyer-friendly market conditions are real, but broad averages are becoming less useful
Aon’s Q1 2026 market overview says capacity and insurer performance kept many placements competitive, but outcomes increasingly depend on product, geography, industry, and risk quality. That distinction matters: a headline soft market does not mean a weak account, U.S.-exposed casualty tower, or geopolitically exposed placement receives the same treatment as a clean property renewal.
Decision Delta - Signal: pricing and coverage are improving where capacity is comfortable. - New vs last issue: this sharpens the renewals story into segmentation rather than generic softening. - Functions affected: pricing, portfolio steering, reinsurance buying, broker challenge. - Action: add a “risk-quality vs market-cycle” bridge to renewal analyses so rate movement is not misread as pure portfolio improvement.
2. Guy Carpenter: alternative capital is moving from tactical capacity to market architecture
Guy Carpenter’s new convergence report frames catastrophe bonds, sidecars, and financial investors as part of the operating architecture of reinsurance, not a peripheral capital sleeve. For actuaries, the issue is not just whether alternative capital is cheaper; it is whether the risk transfer is model-transparent, collateral-aware, and robust under trapped-capital and basis-risk stress.
Decision Delta - Signal: convergence is becoming a structural pricing and capacity force. - New vs last issue: this extends the ILS-capacity theme from transaction flow to capital architecture. - Functions affected: capital modelling, ceded reinsurance, ERM, investor reporting. - Action: when comparing traditional and alternative structures, show attachment probability, expected loss, collateral mechanics, reinstatement assumptions, and model uncertainty side by side.
Read Guy Carpenter’s convergence note
3. Triple-I/Milliman: the U.S. P/C recovery is real, but line-level stress remains hidden under the aggregate
Triple-I’s latest update reports the U.S. P/C industry’s lowest net combined ratio in more than a decade for 2025. The useful actuarial detail is the dispersion: personal auto and homeowners improved materially, workers’ compensation remains strong, but commercial auto and general liability remain above 100 NCR because severity, litigation, and social-inflation pressure have not disappeared.
Decision Delta - Signal: the aggregate combined ratio is no longer the risk story. - New vs last issue: moves from carrier-specific Q1 normalization to market-level underwriting recovery. - Functions affected: planning, reserving, capital, peer benchmarking. - Action: present the industry result as a line-of-business waterfall, not a single-cycle assumption.
Read the Triple-I/Milliman press release
4. NCCI: workers’ comp still looks profitable, but accident-year pressure is visible
NCCI’s 2026 State of the Line guide reports a 2025 calendar-year workers’ compensation combined ratio of 91, while the 2025 accident-year combined ratio is 102. That gap is the actuarial story: prior-year reserve releases can coexist with a current accident year that deserves fresh severity, wage-inflation, medical-cost, and frequency-cycle testing.
Decision Delta - Signal: reserve redundancy is still supporting the calendar-year result. - New vs last issue: adds a workers’ comp reserving lens to the broader P/C profitability discussion. - Functions affected: reserving, pricing, capital, loss-ratio planning. - Action: split workers’ comp dashboards into calendar-year profitability, accident-year adequacy, and reserve-release dependency.
Read NCCI’s 2026 State of the Line guide
5. Aviva: general-insurance growth and ratio improvement give a live peer benchmark
Aviva’s Q1 2026 trading update reports general-insurance premiums of £3.4bn, up 19%, and an undiscounted combined ratio of 94.1%, improved from 96.6% a year earlier. The result is useful less as a headline and more as a benchmark for separating rate, exposure, acquisition, mix, and claims-normalization effects.
Decision Delta - Signal: top-line growth and margin improvement can coexist in this part of the cycle. - New vs last issue: adds a primary-carrier operating result to the reinsurance-heavy market signals. - Functions affected: competitor analysis, planning, pricing, investor messaging. - Action: benchmark Aviva against peers using an adjusted bridge: volume, rate, mix, weather/cat, prior-year development, and expense ratio.
Read Aviva’s Q1 2026 trading update
Deep Dive: Soft markets expose weak actuarial bridges
Softening is not a single assumption. It is a decomposition problem.
The same observed rate change can mean at least four different things: genuine excess capacity, improved risk quality, higher attachment or narrower coverage, or a temporary competitive concession. If the actuarial bridge does not separate those effects, the planning process can accidentally bank market relief that belongs either to underwriting selection or to changed risk transfer terms.
That is especially important now because the signals are pointing in different directions at once. Aon shows broad buyer-friendly conditions; Guy Carpenter shows alternative capital becoming more embedded; NCCI shows workers’ comp calendar-year profitability but accident-year strain; Triple-I shows aggregate P/C recovery with casualty stress still present. The coherent reading is segmentation, not a universal release of risk.
Decision Delta - For pricing: do not flow market softening straight into expected loss ratios without an exposure-quality adjustment. - For reserving: separate accident-year selections from calendar-year reserve-release optics. - For reinsurance: compare ceded-cost reductions with retained volatility, basis risk, and reinstatement exposure. - For capital: scenario-test whether capacity stays available after a meaningful cat or casualty reserve shock.
Research Spotlight
Paper of the Week: Fairness Testing for Algorithmic Pricing
This is the most actuarially relevant arXiv item of the week. The paper argues that the standard fairness audit — regress algorithmic prices on protected attributes and legitimate rating factors, then use ordinary least-squares standard errors — is structurally invalid when the pricing algorithm is deterministic. Applied to quoted premiums from 34 Illinois auto insurers, the corrected method identifies statistically significant conditional demographic parity failures across all 34 insurers, with 16 exceeding a substantive threshold.
The governance implication is sharp: an insurer can have a fairness audit that looks statistically formal but is variance-wrong. That is exactly the kind of failure mode model-risk committees should care about.
From the Journals
Scenario-based CO₂ thresholds for strategic asset allocation and climate stress testing in insurance
The Geneva Papers article extends climate stress testing by asking how insurers can incorporate dynamic management actions and scenario-dependent climate damage forecasts. The practical value is a better bridge between regulatory scenarios, ALM decisions, and explicit transition thresholds.
Generalised Bayesian model averaging for threshold uncertainty in GPD mixture models
The European Actuarial Journal paper tackles a familiar tail-modelling headache: threshold selection in mixture models with generalized Pareto excess losses. Its Bayesian model averaging approach explicitly carries threshold uncertainty instead of pretending one threshold choice is uniquely correct.
From arXiv
The arXiv scraper initially timed out under the default 30-second API limit. Re-running the same scoring function with a longer timeout succeeded and scored 14 relevant papers.
- Fairness Testing for Algorithmic Pricing — corrected inference for deterministic pricing-algorithm audits; directly relevant to insurance pricing fairness governance. Paper
- Neural-Actuarial Longevity Forecasting — anchored LSTMs as model challengers for longevity risk, with SHAP influence mapping, uncertainty calibration, and reverse stress testing. Paper
- Tweedie-based nonparametric estimation for semicontinuous mixed densities — useful for claims or cost data with a mass at zero and a heavy positive component. Paper
- A Beta-GAM Hidden Markov Model for Proportion Time Series — regime-switching beta-GAM structure applied to mortality proportions; relevant for demographic and mortality-state modelling. Paper
- Predictive and Prescriptive AI toward Optimizing Wildfire Suppression — optimization and ML for wildfire response; relevant to cat-risk mitigation, public-sector resilience, and exposure-management thinking. Paper
Practical Takeaways
- Pricing teams: add a deterministic-algorithm fairness-testing check before treating OLS audit p-values as governance evidence.
- Reserving teams: for workers’ comp, show accident-year adequacy separately from prior-year reserve-release benefit.
- Reinsurance buyers: when capacity gets cheaper, re-run retained-volatility and basis-risk scenarios before declaring economic savings.
- Capital / ERM teams: climate and cat scenarios should specify executable management actions, trigger points, liquidity assumptions, and failure modes.
- Model-risk committees: require challenger models to explain not only predictive lift, but anchoring, calibration, uncertainty, and reverse-stress behavior.
What We’re Watching
- Whether mid-year reinsurance renewals confirm segmentation by peril, attachment, and data quality rather than broad softening alone.
- Whether pricing-fairness audit methodology starts moving from legal/compliance language into actuarial model-validation standards.
- Whether U.S. casualty and commercial auto severity pressure begins to offset the aggregate P/C underwriting recovery.
- Whether climate stress testing guidance becomes more explicit about dynamic balance-sheet actions and evidence standards.
Edition 29 • May 15, 2026