Recalibrating the Cat Model After a Record Loss Year: Secondary Perils, AAL Loadings and the Indian Property Book

Swiss Re Institute's sigma 1/2026 landed early in the year with a number every property underwriter should have on a sticky note: secondary perils drove the large majority, on the order of nine in ten, of 2025's global insured natural-catastrophe losses, on a total comfortably above USD 100 billion for another year running. The composition matters more than the headline. Severe convective storm (hail, straight-line wind, tornado) was the single largest contributor at several tens of billions of dollars. The January Los Angeles wildfires pushed the wildfire peril to a record, in the region of USD 40 billion. Global insured flood losses, by contrast, came in below the recent multi-year average despite real damage in parts of Asia and the United States.

Read that last line carefully, because it is where Indian desks get the wrong lesson. A quiet flood year globally does not mean flood is a quiet peril for an Indian property book. It means the loss happened to land elsewhere. Vendor cat models and your own burning-cost views are built to look through single years; the sigma data is a prompt to re-examine the weights, not to chase last year's loss map.

The practitioner question for FY27 is narrow and answerable. When a vendor refreshes its India flood or SCS module, or when your reinsurer's actuary re-derives an average annual loss for your treaty, is the change a genuine reflection of frequency and exposure growth, or is it last year's tail bleeding into this year's mean? You cannot challenge a model number you do not understand. The rest of this post is about understanding it well enough to argue with it.

There is a second-order reason this matters now. Model vendors release new versions on multi-year cycles, and a version released in 2026 will have absorbed the secondary-peril experience of 2023 to 2025 into its calibration. Treaty and primary loadings tend to lag those version changes by a renewal or two as portfolios migrate. So the conversation a desk has at the FY27 renewal is partly about a model vintage decision that was taken upstream, in the vendor's research team, on global data. Knowing that the change you are being quoted may be a vintage artefact rather than a fresh India signal is itself a negotiating position.

Three terms carry most of the weight, so pin them down before a renewal call.

Average annual loss (AAL) is the expected loss per year averaged across the full simulated event set, including the years where nothing happens. It is the pure-risk cost that feeds your technical premium. A model that raises AAL is telling you the long-run expected cost of the peril has gone up.

Event loss and the loss exceedance curve describe the shape, not the average. The model produces a distribution of single-event losses at return periods (the 1-in-100, the 1-in-250). Reinsurers price excess-of-loss layers off this curve. An event loading is the margin a desk adds on top of modelled output for the parts the model under-captures: demand surge after a catastrophe, loss adjustment expense, data gaps in your schedule.

The tail is everything past the return period your capital cares about. Peak perils (the big cyclone making landfall on a metro) dominate the far tail. Secondary perils dominate frequency, the body of the curve, the part that actually erodes your loss ratio year after year.

The recalibration debate is really a debate about which part of the curve moved. Sigma 1/2026 is evidence that the body, driven by hail, convective storm and pluvial flood, is heavier than older model vintages assumed. If a vendor lifts your India SCS or flood AAL while leaving the cyclone tail roughly flat, that is internally consistent with the global signal. If they lift the whole curve uniformly, ask why.

One more distinction earns its keep on a renewal call. AAL drives your pure premium and therefore your proportional and quota-share economics; the return-period points on the exceedance curve drive your excess-of-loss and catastrophe layer pricing. A recalibration can move one without the other. A heavier body (more frequent mid-sized flood and hail) lifts AAL and pressures proportional terms while leaving the far-tail cat layer almost untouched. A heavier tail (a worse modelled landfall) does the reverse. When a counterparty quotes a single blended deterioration, separate it: ask whether your AAL moved, whether your 1-in-100 and 1-in-250 moved, and by how much each. The answer tells you which layer of your programme is actually repricing and which is just being swept along.

Here is the tension a thoughtful desk has to hold. Globally, 2025 was a record secondary-peril year. In India, the April 1, 2026 treaty renewals were among the softest in recent memory. Loss-free excess-of-loss programmes saw meaningful double-digit price reductions reported, on the back of benign domestic loss experience, abundant capacity and strong reinsurer balance sheets. Indian flood and storm did not feature in the global loss tally the way Los Angeles wildfire did.

That divergence is the whole point. The peril mix that recalibrated global models (wildfire, North American SCS) is not the peril mix that hits an Indian commercial property book. Our frequency drivers are monsoon pluvial and riverine flood, urban drainage failure in Mumbai and Chennai, hail and pre-monsoon convective storm across the north and centre, and storm surge on both coasts. A blunt application of the global secondary-peril signal, where roughly nine in ten of the loss sat, to an Indian schedule would be wrong twice over: wrong on which secondary perils, and wrong on the soft pricing reality of the local market.

Do not let a reinsurer use a global record-loss narrative to justify an India loading that the India loss experience does not support. The April renewal showed the local market clearing soft. A peak-peril-led global story is not, by itself, an India reinsurance repricing case.

The correct move is selective. Re-weight toward the secondary perils that genuinely drive Indian frequency, hold the line on the ones that do not, and make the reinsurer show their India-specific event set rather than a borrowed global one. A useful test is to ask which historical Indian events sit in the model's recalibrated set. The 2015 Chennai floods, 2005 and 2017 Mumbai deluges, the recurrent Assam and Bihar riverine flooding and the hailstorm belts of central India are the events that should be re-weighted, not the Los Angeles fire that dominated the global headline. If the reinsurer cannot name the Indian events driving the change, the change is borrowed and you should treat it as such at the table.

Over-fitting is the classic actuarial sin after a loud year: you re-cut the mean to match the most recent observation and quietly bake a tail event into your expected cost. Avoiding it is mostly discipline.

1. Separate frequency change from a single severe event. If your model lifts flood AAL, ask whether the increase reflects more frequent moderate events (a real frequency signal) or one large simulated loss now weighted more heavily. The first justifies a standing loading. The second does not.
2. Blend model output with your own burning cost. Run a credibility blend. A long internal loss history on a stable schedule deserves weight against a freshly recalibrated vendor view. Pure model output on a thin book is fragile; pure experience rating on a growing book misses exposure drift.
3. Trend exposure, not just loss. Much of the apparent secondary-peril increase globally is exposure growth: more value in harm's way, higher rebuild costs. For India, urban warehousing and logistics sprawl into flood-prone peri-urban land is real exposure growth. That belongs in the AAL. General inflation panic does not.
4. Cap the single-year influence. A common guardrail is to limit how much any one accident year can shift a multi-year burning cost. It stops a record year from dominating the mean while still letting a genuine trend come through over three to five years.

The output of this is not one number but a defensible range, with a clear statement of how much is exposure, how much is frequency and how much is conservatism. That range is what you take into the renewal.

A practical guardrail on the credibility blend: where a client's schedule has changed materially (a new flood-exposed warehouse, a relocated process line, a fresh peri-urban site), the historical burning cost on the old schedule is less credible than it looks, because the exposure it describes no longer exists. In those cases the recalibrated model legitimately deserves more weight, and a desk that reflexively defends the old experience figure is arguing from a book that has moved on. Honesty cuts both ways. The discipline is to let the model win where the exposure genuinely changed and to make it earn its increase where the schedule is stable.

On the Indian primary side, most of this lives inside the fire and special perils policy. The Standard Fire and Special Perils cover carries the storm, tempest, flood and inundation (STFI) add-on perils that absorb the bulk of secondary-peril claims. Post de-tariffing, these are priced on the insurer's own view, anchored to IIB-published burning-cost guidance rather than a fixed tariff.

That is where the recalibration actually bites the broker's client. If an insurer re-weights its internal cat view toward flood and convective storm, the STFI component of the fire rate is where you will see it move, often more visibly than the base fire rate. A flood-exposed warehouse near a Chennai or Mumbai drainage line, or a manufacturing shed on reclaimed peri-urban land, is exactly the occupancy a recalibrated model flags.

The practitioner play is to attack the loading with the same granularity the model claims to use. If the model is location-resolved, the loading should be location-defensible. Where the client has put stock on racking above flood level, installed plinth protection or moved critical plant to upper floors, that risk improvement should pull the material damage rate down, recalibration or not. Bring the survey evidence; a generic cat narrative should not survive a specific risk-improvement file.

Secondary-peril recalibration does not stop at the property damage line. The expensive part is usually downstream.

For engineering and project risks, flood and convective storm hit construction sites and erected plant that the standard property cat view may not capture well. A part-built contractors all risks project has no roof, no drainage and a fragile programme; the same return-period flood that a finished building shrugs off can write off a season of work. If the cat model behind a CAR placement is borrowed from a property view, it will understate site exposure during the wet phase. Sequence the works around the monsoon and the modelled frequency risk drops materially.

Business interruption is where the body of the loss curve does its real damage. A moderate flood that causes limited building damage can still take out switchgear, a transformer yard or a ground-floor process line and trigger weeks of downtime. The logistics and warehousing sectors are acutely exposed here: a single inundated distribution centre can stall a regional supply chain even when the structure survives. A frequency-led secondary-peril world raises the BI tail faster than the material-damage tail.

The underwriting implication is to model the contingent and time-element exposure explicitly, not as a percentage add-on to property AAL. Indemnity period, dependency on single locations and reinstatement lead-times for long-delivery plant drive the real number. A recalibration that lifts property frequency without revisiting BI assumptions leaves the worst of the loss un-modelled and the client under-protected on the line that matters most to their cash flow.

Pull this into a desk routine you can run on any property or treaty renewal where a reinsurer or insurer cites model recalibration.

First, separate the global narrative from the India case. Sigma 1/2026 is real and the secondary-peril signal is genuine, but the April 1, 2026 renewal cleared soft on benign Indian experience. Make the counterparty justify an India loading on India data.

Second, decompose any AAL increase into exposure, frequency and conservatism. Ask for the split. A change that is mostly exposure growth (which the client's own balance sheet confirms) is honest and you should accept it. A change that is mostly tail bleed from one record year is over-fitting and you should push back.

Third, take the fight to location level. Recalibrated models claim location resolution, so demand location-resolved loadings. Elevation, watercourse distance, inundation history and risk-improvement evidence are the broker's best ground for arguing the STFI and material-damage rate down.

Fourth, do not let the property conversation skip BI. The frequency drivers that recalibration emphasises (flood, hail, convective storm) hurt most through downtime, switchgear and supply-chain dependency. Price and structure the time-element cover on its own merits.

Done well, recalibration is an opportunity. A desk that can read the loss exceedance curve and tell exposure growth from tail bleed will out-argue one that simply accepts the refreshed AAL and passes the loading through.