The policy is a promise with footnotes
In the previous guide you mapped the P&C landscape — the lines of business and the short-tail-versus-long-tail split. Now we open the document that sits underneath all of it. An insurance policy is simply a contract: the insured pays a premium, and in return the insurer promises to pay for certain losses. But that promise never comes bare. It arrives wrapped in qualifiers that say *what* is covered, *how much*, and *when* — and every one of those qualifiers is a number or a rule that an actuary must eventually price.
A real policy has four moving parts worth naming. The declarations page lists who is insured, the property or activity covered, the limits, the deductible, and the premium — the personalised front sheet. The insuring agreement is the core promise. The exclusions carve out what is *not* covered. And the conditions spell out the duties — report a claim promptly, cooperate, do not lie. As an actuary you rarely litigate this language, but you must read it, because the data you will model later is only ever the losses this language let through.
Perils: what counts as a covered loss
A loss is not covered just because something bad happened. It is covered only if it was caused by a covered peril — the actual cause of loss, the fire or the windstorm or the theft, as opposed to the thing damaged. The peril is the verb of the policy; the property is the noun. A burned kitchen is a covered loss if fire is a covered peril; the very same charred kitchen may be uncovered if the fire was deliberately set by the owner, because arson-by-the-insured is excluded.
Policies define their perils in one of two opposite styles, and the difference is genuinely large. A named-perils policy is a closed list: covered *only* if the cause appears on the list — fire, lightning, hail, and so on. An all-risks (open-perils) policy flips the logic: everything is covered *unless* it is specifically excluded. All-risks is broader and the burden shifts — under named perils the insured must prove the cause is on the list, while under all-risks the insurer must prove the cause is excluded. Two policies can cover the very same house and respond completely differently to the same odd loss.
Exclusions are not the insurer being mean; they are how a policy stays priceable. Flood and earthquake are usually excluded from ordinary home policies precisely because they hit thousands of insureds at once — they break the very pooling that makes insurance work, and so they are sold separately or covered by the state. War, nuclear events, wear and tear, and gradual pollution are excluded for the same family of reasons: either they are not random and independent, or they are not a sudden *fortuitous* event at all, but a certainty dressed up as a risk.
Limits, deductibles, coinsurance — from the policy's side
Even when a loss is covered, the insurer rarely pays the whole thing. Three features decide the gap. The policy limit is the most the insurer will ever pay — a 300,000 dwelling limit means a total loss recovers at most 300,000, no matter that rebuilding costs 380,000. The deductible is the slice the insured keeps on every claim: with a 1,000 deductible, an 8,000 loss pays 7,000 and a 600 loss pays nothing at all. These are the same shapes you met in loss modelling — but here we read them off the contract, not off a severity curve.
Property coinsurance deserves special care, because the word means something different here than in health insurance. In a property policy, coinsurance is a clause that *requires the insured to carry a limit equal to a stated percentage of the property's value* — typically 80%. Insure for less, and a penalty kicks in on partial losses: you recover only in proportion to how badly you under-insured. The clause exists to stop owners gaming the system — without it, everyone would buy a tiny limit, since most losses are partial and small.
Home worth 500,000 coinsurance clause = 80% so required limit = 400,000 Owner actually insured for only 300,000 (under-insured) A partial fire causes a 100,000 loss deductible = 2,000 step 1 coinsurance ratio = carried / required = 300,000 / 400,000 = 0.75 step 2 apply ratio to loss = 0.75 * 100,000 = 75,000 step 3 subtract deductible = 75,000 - 2,000 = 73,000 insurer pays 73,000 owner bears 100,000 - 73,000 = 27,000 (the coinsurance penalty bites)
Higher limits cost more — but not proportionally
Buyers can pick how much limit to carry, especially in liability. So how should a 1,000,000 limit be priced relative to a 100,000 limit? Not at ten times the cost. Most losses are small and fall entirely under 100,000, so the *extra* 900,000 of coverage only ever pays out on rare, severe claims. The expected cost of each higher layer shrinks as you climb. Actuaries capture this with increased limits factors (ILFs): a table of multipliers that scales the premium for a base limit up to any higher limit, growing steadily but always less than proportionally.
A small worked picture. Say the expected limited loss at 100,000 is 5,000, and at 1,000,000 it is 7,500 — the extra 900,000 of coverage adds only 2,500 of expected loss because so little of the distribution reaches that high. The ILF from the 100,000 base to the 1,000,000 limit is then 7,500 ÷ 5,000 = 1.5. You buy ten times the limit for one and a half times the loss cost. Be honest about where this leans hardest: the ratio depends almost entirely on the shape of the *tail* of the severity distribution, the part you have the least data on and the least certainty about.
Occurrence vs claims-made: the clause that bends time
For long-tail liability — the lines where a claim can surface years after the act that caused it — one clause matters more than any limit: the trigger, the rule for which policy year a claim belongs to. The two answers are occurrence and claims-made coverage, and they can assign the very same lawsuit to different years and different insurers.
An occurrence policy responds to losses that *happen* during the policy period, no matter how long afterward the claim is filed. A 2015 occurrence policy must answer a lawsuit served in 2025 if the harmful event took place in 2015. That promise reaches far into the future, which is exactly what makes occurrence coverage so hard to reserve: the insurer of 2015 is still on the hook for claims nobody has reported yet — the IBNR you will study in the reserving rung. A claims-made policy instead responds to claims *first made* during the policy period, whenever the underlying act occurred. The 2025 policy answers the 2025 lawsuit even though the act was in 2015.
Claims-made was invented precisely to tame the long tail. By tying coverage to the year a claim is *reported* rather than the year the harm *occurred*, the insurer shortens its own reporting tail to almost nothing — it knows this year's claims this year — which makes pricing and reserving far steadier. The price of that comfort lands on the policyholder: gaps open up. A retroactive date can exclude old acts, and if you switch insurers or retire, a claim about a past act may fall into no policy at all — which is why claims-made policies sell tail coverage (an extended reporting period) to plug the hole.