U.S. insured losses from severe convective storms exceeded $45 billion in 2025 for the third consecutive year, with a single March outbreak tracking $8 to $10 billion in damages across 26 states and the first confirmed EF-5 tornado in 12 years touching down in North Dakota. Per-event costs ran 31 percent above the prior decade's average, a gap that reflects urban sprawl, construction cost inflation, and a loss amplification cycle that legacy cat models were not designed to capture. Moody's launched its North America Severe Convective Storm HD Models on December 11, 2025, replacing what the company described as tools built on coarse hazard resolution, small event sets, simplified vulnerability curves, and constrained financial modules. For P&C actuaries whose 2026 cat programs were built on legacy SCS model output, the architecture change arrives mid-cycle, when mid-year reinsurance negotiations and regulatory capital reviews are already in motion.

Tracking cat model version updates across the top 15 P&C carriers over multiple product cycles, migration from legacy stochastic to HD model outputs typically shifts SCS probable maximum loss estimates by 15 to 25 percent before any recalibration of reinsurance structure. That gap creates a pricing mismatch in the first renewal cycle after adoption. The pattern observed in wildfire model migrations in California and earthquake model updates following the 2019 Ridgecrest sequence repeats here: the first HD output tends to land higher than legacy estimates, and it takes one to two renewal cycles before treaty structures reoptimize around the revised distribution. The SCS case carries one additional complication. Because severe convective storm is a frequency peril rather than a single-event tail risk, the reinsurance structure that makes sense at a given return period changes materially when the expected frequency and severity distribution itself shifts.

The Legacy Architecture and Its Known Failure Modes

Stochastic catastrophe models for severe convective storms have historically operated from the same design premise: generate a large set of synthetic storm events by sampling from historical frequency and severity distributions, run those events over an exposure database, and aggregate the resulting losses to produce a loss exceedance curve. The framework is sound for perils where the hazard is well-measured, relatively uniform over space, and stable over time. SCS violates all three conditions.

Hail, tornado, and straight-line wind hazards vary sharply over distances measured in tens to hundreds of meters, a spatial scale that coarse hazard grids miss entirely. A legacy model running at one-kilometer or coarser resolution fails to capture the block-by-block intensity gradient that determines whether a property in a dense urban corridor sustains cosmetic hail damage or total roof failure. The event set problem is equally significant. The meaningful U.S. severe convective storm record extends back roughly 70 years in most geographies and is shorter in others, giving stochastic generators a limited basis for tail events. Event sets calibrated on the 1950-2010 record underweight the contribution of derecho events, which can generate losses comparable to small hurricanes but track entirely differently from typical supercell clusters, and they underrepresent the extent of multi-day outbreak clusters whose aggregate damage accumulates across hours-clause boundaries.

The vulnerability side of legacy models carries additional structural limitations. Most legacy SCS models estimate the expected damage ratio for a given hazard intensity, a mean damage ratio approach that discards variance information entirely. That matters for reinsurance pricing because the variance in loss outcomes at a given intensity determines the shape of the loss distribution at attachment. A model that correctly estimates the mean loss but underestimates variance will systematically misprice per-occurrence excess layers relative to aggregate covers. The financial modeling gap compounds both. Policy terms for residential and commercial property now routinely include actual cash value endorsements for roofs, cosmetic damage exclusions for hail, and deductible structures that vary by coverage type and geography. Legacy SCS models either ignored cosmetic damage exclusions or handled them through aggregate rule-of-thumb adjustments applied outside the stochastic engine, overstating gross losses in hail-heavy states where cosmetic exclusions are widespread and understating the frequency of events that breach a given deductible after the exclusion is applied.

Inside the HD Framework: Hazard, Vulnerability, and Financial Precision

The Moody's RMS North America SCS HD Models address each of these gaps through a three-layer architectural change: physics-based hazard simulation at high spatial resolution, vulnerability curves built from actual claims experience rather than engineering judgment, and a financial module that mirrors the policy terms carriers actually write.

The hazard layer uses a physically consistent simulation framework rather than statistical resampling of historical events. Physical simulation can generate rare, coherent event types that are underrepresented in the historical record, including derechos and supershear convective outbreaks, by running the underlying atmospheric dynamics rather than drawing from a bounded empirical sample. It also preserves spatial correlation across the storm footprint at grid resolutions of hundreds of meters, a scale relevant to block-by-block exposure variation in high-density urban markets. The stochastic catalog runs to more than 12 million events across 50,000 simulated years, with 16 distinct hail size categories providing the granularity needed to distinguish cosmetic damage from structural failure at the individual property level. Derechos, which legacy models often treated as anomalous or excluded from event sets entirely, receive explicit simulation treatment because their loss dynamics, specifically the extent of straight-line wind damage across a wide geographic corridor, differ from supercell patterns in ways that matter for aggregate program structuring.

The vulnerability layer is calibrated against $55 billion in location- and policy-level claims data, described by Moody's as the largest such library assembled for SCS model development. The output is more than 2,700 property damage curves covering residential, commercial, automobile, and specialty occupancies including renewable energy assets. Critically, these curves represent the full loss distribution for a given hazard intensity rather than the mean alone. The roof age, roof geometry, and construction quality modifiers that drive most of the variance in observed hail loss outcomes are explicit inputs, not post-processing corrections. Automobile coverage and renewable energy exposure, both growing fast in hail-prone corridors, get their own curve families rather than being proxied from residential building data.

The financial module introduces explicit deductible structures, cosmetic damage exclusion clauses, flexible hours clauses for clustered events, and a parameterized post-event loss amplification function into the core stochastic engine. Post-event loss amplification has been a consistent source of model miss in recent large SCS years. The July 2023 North Texas hail outbreak generated approximately $8 billion in insured losses in part because roofing contractor capacity in the Dallas-Fort Worth corridor was saturated for months after the event, inflating ultimate losses well beyond the initial physical damage estimate. Embedding this function in the financial module rather than applying it as a post-processing factor allows it to interact correctly with the simulated hazard magnitude and the portfolio-level demand surge exposure, rather than being applied as a flat percentage adjustment that ignores event size and geographic concentration.

Why Per-Event SCS Costs Have Risen 31 Percent

The 31 percent increase in per-event SCS costs relative to the prior decade average is not principally a model-measurement artifact. Three structural factors are driving it, and all three interact with model calibration in ways that matter for 2026 budgeting.

Urban infill in the Tornado Alley and Dixie Alley corridors has placed significantly more insured value in hail and tornado exposure zones over the past 15 years. The Dallas-Fort Worth metro, the fastest-growing large urban area in the United States by absolute population gain over the 2010-2025 period, sits in one of the highest hail frequency corridors in the country. The insured value density along the I-35 corridor from Dallas to Oklahoma City has increased materially since most legacy SCS models were last fully recalibrated. Coarse geographic resolution prevents legacy hazard grids from capturing the block-by-block exposure intensification that has occurred through infill development and vertical construction in previously low-density zones.

Construction cost inflation since 2021 has been substantial. Residential replacement costs in hail-heavy states increased 25 to 40 percent between 2020 and 2024, driven by lumber, steel, roofing labor, and material costs. Legacy vulnerability curves calibrated on pre-2021 claims data embed lower replacement costs than the current exposure base, understating expected loss at a given damage ratio. Some carriers have addressed this through blanket inflationary adjustments to their exposure databases, but these adjustments interact unpredictably with legacy vulnerability curves that were not designed to be inflation-indexed. The result is a systematic underestimate of the gross loss for high-intensity events, precisely the events where the mismatch between modeled and actual replacement cost is largest.

Post-event loss amplification has intensified in a way that correlates with event size. The concentration of SCS losses in a smaller number of high-damage events, rather than a broader distribution of moderate events, creates demand surge pressure that compounds the underlying physical loss. The March 2025 outbreak's $8 to $10 billion in insured losses was the largest SCS event in the modern record, and the contractor and adjuster capacity strain it created in the affected 26-state corridor carried into Q2 2025, inflating ALAE and ULAE on claims that were not initially extraordinary in severity. A model that does not embed post-event loss amplification as a function of event size will consistently understate the ultimate loss for the events that matter most to aggregate cat programs.

PML Migration: From Legacy Output to HD Estimates

The practical question for P&C actuaries running 2026 mid-year cat budget reviews is what the HD model output implies for existing SCS PML estimates. The directional answer from cat model version migrations across multiple product cycles is consistent even if the magnitude varies by carrier: HD output tends to land higher than legacy output, and the gap is largest at shorter return periods rather than at the extreme tail.

That asymmetry has a structural explanation. Legacy SCS models, because of their coarse hazard resolution and simplified vulnerability curves, tend to understate the frequency of moderate-severity events while not systematically misestimating the extreme tail, where the physical limits on convective storm intensity impose natural constraints on maximum loss. The HD model, by correctly representing the full vulnerability distribution and capturing spatial clustering of moderate-intensity hail events over urban exposure concentrations, raises the expected frequency of events in the 1-in-10 to 1-in-50 return period range more than it raises the 1-in-250 or 1-in-500 tail estimates. The improvement in the middle of the distribution, not the tail, drives the aggregate cat budget revision.

The implication for cat budget reviews is that the SCS aggregate probable maximum loss figures feeding into combined catastrophe load calculations need to be revisited. The standard approach of using vendor model output as a starting point and applying credibility adjustments to incorporate carrier experience may understate the shift if credibility weights were calibrated on legacy model output. Carriers whose portfolios have significant residential property concentration in Texas, Oklahoma, Kansas, Nebraska, or the upper Midwest should expect HD model output to show meaningfully higher aggregate exposure in the 1-in-20 to 1-in-50 range. That is the range that typically determines aggregate cover attachment points, and it is where the structural upgrade in hazard resolution and vulnerability representation has the most direct effect on the modeled loss distribution.

Carriers that have adopted HD model output for wildfire following the California CDI certification process in 2025 have already worked through the model migration workflow. That process required rate exhibit documentation, internal model governance sign-off, and reinsurance submission updates that are directly transferable to the SCS migration. The wildfire precedent also established a market expectation that carriers adopting certified or current-generation models will update their reinsurance submissions accordingly, not simply carry forward legacy PML figures while running HD output internally.

Reinsurance Purchasing: Attachment Points, Layer Sizing, and Aggregate Structures

The SCS reinsurance purchasing decision is structurally different from a hurricane or earthquake program because the underlying peril is frequency-driven. A carrier with material SCS exposure in the Central Plains states can expect 10 to 20 separate events per year that generate loss notifications, of which three to five will produce losses above a routine-loss threshold. The reinsurance structure that makes sense for this exposure profile is categorically different from a catastrophe excess-of-loss tower designed for an infrequent, high-severity peril. Aggregate stop-loss covers, per-occurrence protections with low attachments, and quota shares are all more commonly used in SCS programs than in hurricane programs, and each interacts differently with a shift in the frequency-severity joint distribution.

The migration to HD model output affects the SCS purchasing decision on three dimensions. First, if the HD estimate of the 1-in-10 aggregate return period is materially higher than the legacy estimate, the optimal attachment point for an aggregate cover must be recalibrated upward to maintain the same protection level, or the premium load will increase because the cover now attaches with higher frequency than intended. Carriers that price their reinsurance programs based on legacy SCS PML and then adopt HD output are effectively discovering mid-cycle that their aggregate protection is thinner than modeled. Second, the layer sizing above the attachment point is sensitive to the shape of the revised loss distribution. If HD output narrows the dispersion of aggregate outcomes by better modeling the frequency-severity joint distribution, a given layer provides more stable protection and warrants different pricing than legacy output would imply. Third, carriers that relied on the legacy model's frequency underestimate to justify a minimal aggregate program may need to reassess whether a per-occurrence structure, a hybrid aggregate-occurrence structure, or a broad corridor cover is now more cost-effective given the revised distribution.

The mid-year 2026 reinsurance market context creates a practical opportunity. Property cat rates fell 15 to 20 percent at June 2026 renewals, driven by record dedicated reinsurance capital and a below-normal hurricane forecast. Carriers that adopt HD model output ahead of mid-year negotiations have an informational advantage: they can request attachment point adjustments supported by the revised distribution and can potentially extend aggregate program limits at softer pricing rather than accepting the legacy structure unchanged. The methodology for translating a reinsurance rate change into a cat load adjustment in primary rate filings is well-established, but that workflow assumes the underlying PML is correctly stated. A carrier that has not updated its SCS PML to reflect HD output is making both errors simultaneously: misstated PML and a reinsurance cost allocation based on the misstated distribution.

The NAIC RBC Signal: SCS Capital Charges Take Shape

The NAIC Property and Casualty Risk-Based Capital Working Group and its Catastrophe Risk Subgroup have been working toward a formal SCS capital charge for several years. The current state of the Rcat formula is that severe convective storm is reported in the hurricane and earthquake disclosure framework but classified as informational, meaning it does not contribute to the Rcat component of the P&C RBC formula the way hurricane and earthquake do. The Catastrophe Risk Subgroup adopted a formal wildfire Rcat charge (Proposal 2025-20-CR) at Spring 2026, adding wildfire to the formula alongside hurricane and earthquake, and the July 9, 2026 meeting is expected to advance the parallel severe convective storm analysis. The NAIC's consolidation of its three cat-related oversight bodies into a single Natural Catastrophe Risk and Resilience Task Force in 2026 signals that SCS is on the same regulatory trajectory as wildfire, not a deferred question.

When the SCS Rcat charge becomes formal, the capital implications for carriers with significant Central Plains, Texas, or Midwest residential property concentration will be substantial. The wildfire charge adopted in Spring 2026 required carriers to use one of four NAIC-approved vendor models to estimate their net 1-in-100 PML, with the capital charge calculated as a factor applied to that estimate. If the SCS charge follows the same structure, and the approved vendor list includes the Moody's RMS HD Models, early adopters who have recalibrated their SCS PML upward to match HD output may face a higher near-term RBC capital charge than carriers still running legacy model output.

The strategic calculation is not straightforward. Adopting HD output and recalibrating PML upward reduces model risk and provides a more defensible capital position in a downside scenario, but it may also front-load the regulatory capital charge that arrives when the formal SCS Rcat requirement takes effect. Actuaries supporting their carrier's capital planning function should model both the current-state RBC impact of adopting HD output and the expected impact when the formal charge applies, rather than treating these as sequential decisions. The carrier that defers HD adoption to avoid a near-term RBC charge may find the deferred transition coincides with the regulatory deadline, compressing the migration timeline and eliminating any first-mover informational advantage.

Communicating the Model Migration to Rating Agencies and Reinsurers

Rating agencies and reinsurance counterparties both care about model change, but for different reasons. Rating agencies, particularly AM Best and S&P Global, focus on whether a carrier's cat risk management program remains internally consistent after a model change and whether the revised PML estimates are credibly supported by the new methodology. Reinsurers focus primarily on the quality and recency of the cat submission they use to price treaty exposure.

AM Best's P&C RBC cat model guidelines require that changes in modeled PML above a threshold relative to prior-year estimates be disclosed and explained in the actuarial opinion on risk-based capital. If an HD model migration produces a 20 percent increase in the 1-in-100 SCS PML, that shift must be explained in the context of the model architecture change, not simply absorbed as unexplained year-over-year variance in the cat load. Agencies look for evidence that the carrier tested the new model output against recent event experience, understands the source of the PML increase, and has updated its reinsurance structure and retention decisions accordingly. A carrier that adopts HD output but does not revise its reinsurance program structure will face a harder explanation than one that presents the model migration and the treaty changes as a coordinated response.

For reinsurance submissions, the practical guidance is to lead with the architecture change rather than the PML number. A reinsurance counterparty reviewing a cat submission that shows a 20 percent PML increase without explanation will price the uncertainty conservatively. A submission that walks through the HD model's calibration methodology, the specific vulnerability curve improvements relevant to the carrier's geographic concentration, and the expected effect on the aggregate distribution gives the underwriter a framework to assess the quality of the change. The 2026 mid-year renewal cycle is the right time to initiate these conversations, even for carriers that will not formally migrate until the 2027 annual renewal. Reinsurance underwriters who understand that the next cat submission will reflect HD model output can begin incorporating that expectation into renewal pricing, rather than encountering a step change in PML at the next renewal date without context.

The Forward Look: Frequency Perils and Model Adequacy

The Moody's RMS SCS HD Model launch is part of a broader pattern of model investment in secondary perils that dominated 92 percent of global insured nat cat losses in 2025. Verisk's Synergy Studio cloud-native platform and the wildfire model certifications in California represent the same underlying market signal: the industry's historical model investment in hurricane and earthquake has not matched where insured losses actually concentrate in a world of persistent SCS, wildfire, and flood frequency. The overlapping peril problem, where SCS, hail, and wind events interact across the same geographic corridors as flood and wildfire exposure, compounds the model adequacy question because legacy models treat these perils as independent when the loss drivers are often correlated.

The per-event cost trend that has driven SCS losses 31 percent above the prior decade average is not a one-time recalibration event. Urban exposure density in high-hail corridors will continue to increase as population growth in the Sunbelt and Plains states outpaces the national average. Replacement cost levels, while moderating from the 2021-2023 inflation spike, have not reverted to pre-pandemic baselines. Post-event loss amplification risk grows with each consecutive high-severity event season, as contractor and adjuster capacity is stretched across an increasingly large event footprint. The HD model is not a response to an anomalous 2025. It is a response to a structural shift in where SCS losses concentrate and how large individual events can become, a shift that has been building for a decade and that legacy model architecture was not designed to capture.

The immediate action list for P&C actuaries in the 2026 mid-year budget cycle: re-run the SCS aggregate PML under HD model output and compare it to the current cat load assumptions in the pricing basis; assess whether existing reinsurance attachment points remain appropriate given the revised distribution; confirm that the company's cat submission to reinsurers accurately represents the current model generation; and document the analysis in a form that supports both AM Best disclosure and the actuarial opinion on the cat component of RBC. Carriers that built their 2026 cat programs on legacy model output are not necessarily wrong about last year's losses. They are likely wrong about next year's distribution, and that gap has a measurable cost at attachment.

Further Reading

Catastrophe Modeling Severe Convective Storms Reinsurance

Sources

Stay ahead with daily actuarial intelligence - news, analysis, and career insights delivered free.

Subscribe to Actuary Brew Browse All Insights