· Physical climate risk · Week 5
Case Insurance
Northwood Capital · physical climate risk

Portfolio edition: Kith Climate cohort case, fictional client. Prepared for Sarah Mitchell, COO. Report date 16 October 2026.

What physical climate risk could cost Northwood before each exit, and what to do about it

All ten portfolio companies. Every figure is a range (low, central, high); the Case switch at the top picks which one is written large, and the Insurance switch moves what the companies keep, the premium path and the lender view between today's terms and a market that has pulled back. Gross loss does not change with insurance.

1 · HazardWhat happensA named event at 1-in-10, 1-in-50 or 1-in-100 (what these mean), in one of six regions
2 · ExposureWhat is in the way
3 · VulnerabilityWhat breaksDays down, supplier outages, stock and substitution, flood depth-damage, heat on outdoor work
4 · DollarsWhat it costsGross loss, what insurance pays, what the company keeps, the next renewal, covenant headroom, value at exit

Reading 1-in-10, 1-in-50 and 1-in-100

These describe how big an event is, not when it comes. A 1-in-50 event is one so severe that an event at least that bad has about a 2% chance of happening in any single year. It is not a schedule: it can happen twice in a decade, or not in a century. Over several years the chances add up, which is why the hold and the buyer's years matter.

Where the event losses come from, and what the labels mean here

This is a bottom-up stress model, not a probabilistic catastrophe model. A hazard counts in a region only where FEMA's National Risk Index gives it a meaningful yearly frequency at a county with a site or supplier. The 1-in-50 setting is built from assumed downtime ranges by facility type and hazard, with 1-in-10 and 1-in-100 at 0.4 and 1.6 times those days. Flood damage uses FEMA Hazus depth-damage curves, wind uses stated damage ratios, and supplier outages run 1 to 26 weeks before stock and alternates apply. So "1-in-50" labels a severity chosen to match that return period, not an intensity drawn from a hazard catalogue, and the chances above describe what the label means.

Footprints are simple and stated. A hurricane hits every site in its region at once, which overstates the coastal scenarios for companies spread along the coast. A river flood hits each company's worst county or its single worst supplier. A convective storm hits one site per company at 1-in-50. Where the flood map and the company disagree, or a point has only a city, flood status is carried as a range.

Where to look: each company in each scenario, on Northwood's own severity scale

Each scenario cell is the share of the company's EBITDA it keeps after insurance, plus the next renewal step (the measure the engine tests covenants on). Insurance terms are estimated from typical mid-market cover until the companies' schedules arrive, so these colours depend on that estimate. The Expected a year column is gross expected loss ÷ EBITDA, before insurance. Bands are Northwood's severity scale: Low under 5%, Medium 5 to 15%, High 15 to 30%, Critical over 30%. Severity is only half of Northwood's framework: a 1-in-50 event has under a 10% chance over any hold here, which the framework rates Remote, so by its own priority matrix Critical severity ranks as High priority and High as Medium. Switch to priority below. Click a cell to open that scenario at that company; click a dash to see that company's own worst event.

Map

Where the sites and suppliers are, and what each one puts at risk

Every own site, supplier and managed property across the ten companies. Shape says whose it is, size is the 1-in-50 gross loss of the dependencies at that point, and colour is Northwood severity of its worst one (or the company). Pick a scenario to light up its footprint. Click a point for its detail.

Scenarios

Seven scenarios, run through from the event to the buyer's years

Each preset is a fixed setting of the same precomputed grid the builder reads. Step through it: the event, the days down, the cash squeeze, what insurance pays, the next renewal, the covenant test, the exit and the buyer's years. The Case and Insurance switches apply throughout.

Why these seven, and how every company is covered

Insurance

Same event, different terms: what a pull-back would cost

What the companies keep, today and after a pull-back

Each preset at the selected case. Blue: kept on today's terms. Orange: kept after a pull-back. The gap is the insurance terms, not the hazard.

Premiums: today, at exit and on the buyer's horizon

Premium above a 3% non-climate trend, a year, at the selected case. Hardened market adds higher retained deductibles.

Exit timing

How much of the cost Northwood carries, and how much reaches it only through the exit price

How much of the recurring cost would a buyer take off the price?

Part of the recurring cost is already in the exit-year earnings a buyer pays a multiple on, so it is priced whatever happens. What is still at stake is whether a buyer expects it to keep growing (and discounts the price for that), and whether Northwood brings it down before the sale. Our working view, with no source behind it: a buyer is more likely to discount visible lines, such as insurance creep, than modelled ones, such as heat productivity, which appear in no accounts. Below, the price at risk is split that way. Pick a share to see the range.

Northwood's hold (report date to exit)The buyer's years (5 after exit, assumed)

Outlook

How the climate at each company's sites is projected to change, and what more frequent or severe events would cost

Two kinds of change, shown separately because the evidence behind them differs. Slow trends in heat and rainfall have county projections; of these, only days at or above 95°F feed a priced cost (heat on outdoor work on the buyer's horizon), and the rest are shown for context. Changes in how often storms, floods and droughts strike have no county projection the Week 5 work could use, so the second panel lets you set your own and see what it would cost.

1 · Projected trends at each company's own sites

Why these measures
Illustrative what-if

2 · If events became more frequent or more severe

Pick the hazards and set your own uplift. The page multiplies those hazards' part of expected loss by (1 + frequency) × (1 + severity), evenly at every return period, and leaves the rest on today's climate. Every hazard starts ticked; the share beside each is its part of expected loss today, so you can see where an uplift matters. The sliders run to a doubling of frequency and half as severe again: wide enough for a stress case, and the page suggests no value. It is a sensitivity of the stated answer, not a forecast.

More frequent: 0%
More severe: 0%
What to do

Decisions for Northwood, with owners, costs and the break-even

Each lever was re-run through the same engine. Value protected is the expected loss it avoids over the part of the hold after it is in place, plus the recurring cost it avoids times the exit multiple if it is in place before exit (the same upper-bound basis as Exit timing). The verdict follows a stated rule on the central case, and the costs are stated judgment until quotes replace them. The break-even line asks a narrower question: would the loss avoided in one 1-in-50 event alone repay the cost?

Every lever, as a decision card

The full lever table, with the verdict rule

Appendix A · Evidence

Why the answer holds, and where it is weakest

Supporting material for the views above: the companies' own past events replayed, the assumptions that move the answer most, and the register every figure is built from.

An average year against one severe event

Expected loss a year (blue) against the company's largest 1-in-50 preset (orange), gross, at the selected case. Averages understate what one company must absorb in one bad year.

The companies' own past events, replayed

Each event the companies reported for 2023 to 2026, run through the engine with its actual sites, hazard and length at 1-in-10 severity, against what the company reported. The shaded band is a pass (0.5 to 2 times). The first run passed 5 of 9. Two downtime assumptions (food-plant winter, clinic hurricane) were then narrowed using these same events, and two events were set aside by decision: Keystone's own session count and reported revenue contradict each other, and Ridgeline's reported loss was project work deferred, not a closure; all 7 tested now pass. That is a fit to the evidence available, not an independent validation, and it covers small recent events, not 1-in-50 losses. The replay compares revenue lost or delayed, most of which comes back, while the scenarios report loss that does not. So Pinnacle's $3.20M from two 2025 storms (modelled $2.96M at 1-in-10) is not the same measure as its $3.53M loss in a 1-in-50 Gulf landfall:

The six assumptions Week 5 tested

Each assumption moved end to end with everything else central; bars show the change in the answer it moves, as a share of that answer's central value.

The wider test, against every headline (November 2026)

Week 5 tested the six assumptions above. The portfolio review found others that drive the headlines but had never been moved, so each was re-run through the same frozen engine, one at a time, at its low and its high value with everything else central. The base run reproduces the stored figures exactly. Bars show the full swing as a share of the central figure; pick a headline.

Pinnacle's premium growth is company data (12% a year from its own premium history), so it is moved across its broker's stated 10 to 15% instead of a ledger range. Not moved: the 3% baseline premium trend (no range), the business-interruption waiting period (no range), the pull-back premium step (it only moves the after-pull-back view), and the other untested assumptions listed in Appendix B.

Vulnerability register: what each company depends on

One row per dependency, worst plausible hazard at 1-in-50, today's market. The full 223 rows, with search and sort, are on page 05.

Calibration detail and open data requests
Appendix B · how it is built

The system that makes each number defensible

Every figure on this page comes out of one chain: three kinds of input, a register of what each company depends on, one engine, and a frozen set of stored results. The work passed three signed gates and eight automated checks; this appendix says which controls cover which step, and where there is none. Click any step to see what goes in, what happens, its controls, and the assumptions that enter there. Amber pins count the flagged assumptions at that step.

Assumptions

The view starts on the flagged ones, the assumptions a reviewer should look at first: tested: moves the answer moved end to end, and it shifts a headline figure by 5% or more; awaiting data a data request to the company is open; stand-in estimated by rule until the company's own data arrives; decided under delegation set by the builder under Daniel's delegation, not by him directly. Pick a step to see the rest; click a row for its basis, use and status.

Rebuild the headline figures from the stored grid

1 · Expected loss a year

For each plausible hazard and region, with l10, l50, l100 = gross loss at 1-in-10, 1-in-50, 1-in-100: annual = (0.10 − 0.02) × (l10 + l50) / 2 + (0.02 − 0.01) × (l50 + l100) / 2 + 0.01 × l100 Expected loss a year = sum over every plausible hazard and region; over the hold = × hold years

2 · The "do nothing" stack over the hold

stack = Gulf landfall (1-in-50) + Ohio Valley winter storm (1-in-50) + recurring cost a year × hold years (a register row hit by both events is counted once, so a company's stack can sit slightly below the plain sum)

3 · Recurring cost at exit, and what a buyer may take off the price

recurring a year = premium above a 3% trend + heat productivity + cooling + retained deductibles (hardened market only) price at risk (upper bound) = recurring a year × the company's exit multiple; the page also shows half and none

4 · New arithmetic in this portfolio edition

5 · Quality checks and gates