Prices that breathe: hard and soft markets
You have just learned to keep score with the combined ratio. Now zoom out from a single year to a couple of decades and a strange rhythm appears: the whole non-life market does not sit at a steady price. It swings. For a stretch of years insurers cut prices, relax terms, and chase market share — a soft market — and combined ratios drift upward toward and past 100%. Then something snaps, prices leap, coverage tightens — a hard market — and for a few years insurers make handsome underwriting profits. Soft, hard, soft, hard. This long, recurring swing in price and profitability is the underwriting cycle.
Get the labels the right way round, because the words feel backwards at first. A *hard* market is hard for the buyer: cover is expensive and scarce, insurers are picky, and a renewal quote can double. A *soft* market is soft for the buyer: cover is cheap and plentiful, and underwriters will write almost anything to keep the premium flowing. The hardness or softness describes the price and the appetite of the sellers, not the demand of the buyers — demand for auto and home insurance barely moves, yet the price riding on top of it heaves up and down.
Why the cycle refuses to die
Here is the honest puzzle: in a textbook competitive market, persistent profits should attract entrants and persistent losses should drive exits until price settles. So why does P&C insurance instead oscillate, decade after decade? No single cause; rather a handful of forces that reinforce each other. The deepest is a problem you already half-know from reserving: in a long-tailed line, you do not know how good or bad a year's business really was until claims have run off, years later. You are flying with a fogged windscreen, pricing today on results you will only confirm long after.
Now layer on the other forces. Capital matters: a big loss or a market crash dents insurers' surplus, and with less capital to support risk they must charge more — supply contracts and the market hardens. As profits return, capital rebuilds, new players pile in attracted by the good years, competition resumes, and prices soften again. Behaviour and accounting amplify it: when reserves on the foggy past years finally prove inadequate, everyone discovers their losses *together*, panics *together*, and raises prices *together*. The cycle, in short, is what you get when many firms make slow, lagged, herd-like decisions about a product whose true cost is only revealed long after it is sold.
Catastrophe risk: when many losses arrive at once
Recall the very first idea of this whole ladder: insurance works because pooling many *independent* risks lets the law of large numbers tame the average. A hurricane breaks that promise. When a single storm floods ten thousand homes in one afternoon, those ten thousand losses are not independent — they are one event wearing ten thousand masks. The pool does not diversify them away, because the same peril strikes them all together. That is catastrophe risk: the danger of a single rare event — hurricane, earthquake, wildfire, flood, and increasingly cyber and pandemic — producing an enormous number of correlated claims at the same instant.
Catastrophe losses are rare but huge — the very definition of a heavy tail. Ordinary years have none; a bad year can wipe out a decade of profit. This breaks the ratemaking habit you are about to learn, where you average past losses to predict the future. If a region suffers a 1-in-100-year quake roughly once a century, then in any ten years of data you most likely see *zero* such quakes — averaging your history would set the price at nearly nothing, which is catastrophically (literally) wrong. You cannot price a rare event from a short sample of its absence. So actuaries reach for a different tool: a model of the physical hazard itself, run thousands of times.
Catastrophe modelling, at a high level
A catastrophe model is a simulation of disasters that have not happened yet. Rather than averaging the thin past, it builds a giant library of plausible synthetic events and asks: if *this* particular storm hit *your* particular book of business, what would you pay? It is usually assembled from four linked modules, each answering one question.
- Hazard — where, how often, and how strong? Generate tens of thousands of synthetic events (storm tracks, fault ruptures) consistent with the science and the historical record, each with an annual probability.
- Exposure — what do you insure, and exactly where? Pin every insured building to its location, construction, height, and value — the geography of your own portfolio.
- Vulnerability — given that intensity at that spot, how much damage? Damage functions turn wind speed or ground shaking at each location into a percentage of value destroyed.
- Financial — after deductibles, limits, and reinsurance, what does the insurer actually pay? Apply the policy terms you met earlier to convert ground-up damage into the insurer's net loss for each event.
Run all those events through all four modules and you get a loss for each, weighted by its probability. Sort the results and you have the exceedance probability curve — the chance of losing at least X in a year. From it actuaries read off the two numbers that govern catastrophe capital and reinsurance buying. The first is the AAL, the average annual loss, which feeds the catastrophe loading in the price. The second is a tail measure such as a high-percentile loss — closely related to the tail value-at-risk you will meet later, the *average* loss beyond a chosen probability. Note the honest distinction: a plain percentile (the loss you exceed only 1% of the time) ignores how bad the worst 1% actually gets; the tail-average looks *into* that tail. For catastrophes, where the disaster lives precisely in the far tail, that difference is not academic.
Money that comes back: salvage and subrogation
Not every dollar that flows out as a claim stays out. After a loss is paid, two channels often send money *back* to the insurer, together known as salvage and subrogation — recoveries that reduce the net cost of claims. Salvage is what is left of the damaged thing. If an insurer pays a policyholder the full value for a car written off in a crash, the wreck now belongs to the insurer; selling it for scrap or parts recovers, say, a few hundred dollars. Subrogation is the insurer stepping into the policyholder's shoes to pursue whoever actually caused the loss. If a careless contractor flooded your home, your insurer pays you promptly, then sues the contractor (or bills *their* insurer) to recover what it paid.
Both rest on the principle of indemnity: insurance should restore you to where you were, not leave you better off, so you cannot keep both the payout and the wreck, nor collect twice for the same loss. For the actuary the practical point is timing and netting. Recoveries trickle in *after* the claim is paid — a wreck sold next month, a subrogation suit settled two years on — so losses can be measured gross (before recoveries) or net (after). Mixing the two corrupts everything downstream.
One auto total-loss claim Gross loss paid to policyholder ............ 20,000 less Salvage (wreck sold for parts) ........ -3,000 less Subrogation (recovered from at-fault) . -7,000 ------------------------------------------------------ Net loss to the insurer .................... 10,000 Gross loss ratio (premium 16,000): 20,000 / 16,000 = 125% Net loss ratio (premium 16,000): 10,000 / 16,000 = 62% Same claim. Two very different stories.
Setting the stage for reserving and ratemaking
Stand back and see how this rung's vocabulary now points forward. Catastrophes are the reason a single year's losses are not enough to set a price; salvage and subrogation are the reason you must say whether you mean gross or net; the cycle is the reason today's price level is itself an assumption rather than a fact. All three feed the same two crafts you are about to learn. In ratemaking you will turn losses into a pure premium per exposure — and you will have to load it for catastrophes the data did not happen to contain, net it correctly for recoveries, and judge where in the cycle you are pricing.
In reserving the same forces reappear from the liability side. After a big catastrophe an insurer must estimate, fast, how much it will owe on claims still being reported and still developing — losses that will trickle through a development triangle for years, net of recoveries that have not yet arrived. And because the insurer can shed the worst of the tail to others, the picture is rarely the full gross figure: a layer of catastrophe loss is passed to reinsurers through catastrophe excess-of-loss cover, so that one storm dents the surplus rather than shattering it. Hold these three ideas — cycle, catastrophe, recoveries — and the twin crafts ahead will feel less like new machinery and more like the natural questions this world was always going to ask.