From claim cost to a posted premium
In the previous guides you learned that a [[health-claim-cost|health claim cost]] is frequency times severity, dragged forward each year by medical trend. That gives you the expected *claims* for a member — but a member does not pay claims, a member pays a premium. This guide is the bridge: how the noisy, morbidity-driven cost you have been studying turns into a single number someone actually pays each month, and how the insurer keeps honest books for claims it has incurred but not yet paid. Two activities, tightly linked: rating (setting the price up front) and reserving (measuring what you already owe).
The skeleton is the same gross-premium logic you met on the premiums rung, just sped up to a yearly cadence. Start with the expected claim cost. Add the cost of running the plan — administration, broker commissions, claim-handling staff, and the legally required margin to feed reserves and capital. Layer in the expense and profit loadings. Because most health cover is one year long, the time value of money barely bends the answer — a stark contrast with the decades of discounting that dominate life pricing. The premium is mostly *next year's claims plus the cost of paying them*, projected one year forward by trend.
Building a monthly health premium (one member): projected claim cost next year = $4,536 (this year's $4,200 grown 8% by trend) administration & claim handling = + $480 broker commission = + $250 margin for reserves & profit = + $360 -------------------------------------------------- annual gross premium = $5,626 monthly premium = 5,626 / 12 = $469 / month Claims are ~81% of the premium -> target loss ratio about 0.81
Community rating vs experience rating — and the fairness fight
The premium above was for one member, but the deepest question in health pricing is *whose* claims it should reflect. The two poles are captured by [[community-vs-experience-rating|community rating versus experience rating]]. Under community rating, everyone in a defined pool pays the same rate (perhaps varied only by a few allowed factors like age or region) regardless of their own health. Under experience rating, a group's — or even an individual's — premium is pulled toward *their own* claims history: a company whose staff ran up heavy claims last year pays more next year. The same expected-cost arithmetic underlies both; they differ only in how finely they slice the pool.
Now the fairness fight, because both poles are "fair" — by different definitions. Experience rating honours *actuarial fairness*: you pay for the risk you bring, the principle of rate equity that keeps a system from quietly forcing the healthy to subsidise the sick. Community rating honours *social solidarity*: health is partly a lottery of birth and luck, so a diabetic and a marathon runner pay alike and the burden is shared. Neither is wrong; they optimise different things. This is why health insurance is so heavily regulated — laws often *ban* charging more for pre-existing conditions, deliberately choosing solidarity over actuarial fairness. The actuary's job is not to win that argument but to price honestly *inside whichever rule applies*.
In practice almost no plan sits at a pure pole; most blend the two through [[credibility-premium|credibility]]. A group's own experience is partly believable and partly noise, so the premium is a weighted average: credibility times the group's own rate, plus one-minus-credibility times the community (manual) rate. A 10,000-life employer gets heavy weight on its own data; a 12-person startup gets almost none, because one unlucky case would swamp its tiny sample. Credibility is simply the data telling you how much of "their own number" you are statistically allowed to trust — the same large-numbers idea from pooling, applied to a single client.
Antiselection: the death spiral that haunts the pool
Choosing how finely to rate is not just a fairness question — it is a survival question, because of [[antiselection-in-health|antiselection]]. This is the health-specific face of the adverse selection you met long ago: the people most eager to buy, or to keep, generous coverage are disproportionately those who expect to use it. If a pool is community-rated and joining is voluntary, the healthy look at the average-priced premium, decide it is a bad deal *for them*, and walk away. Their leaving raises the average claim cost of who remains, which raises next year's premium, which chases out the next-healthiest tier — and round it goes.
Because regulators often forbid the obvious fix — charging the sick more — insurers and policymakers fight antiselection with structural tools instead. Mandatory enrolment removes the choice to opt out, so the healthy cannot flee. Open-enrolment *windows* and waiting periods stop people from buying coverage only on the way to the operating theatre. Pre-existing-condition exclusions, where still allowed, fence off claims that were already brewing at signing. Even employer group plans lean on this: because everyone at a firm is enrolled together for a job, not for their health, the group is naturally close to a random cross-section — antiselection defanged by bundling.
Reserves and IBNR: paying for claims you cannot yet see
Switch now from pricing the future to measuring the past. When a member gets sick in December, the insurer *owes* that money the moment care happens — but the bill might not arrive until February, and might not be paid until April. At any year-end the company is therefore sitting on a real liability for care already delivered but not yet settled. The estimate of that liability is a [[health-reserves|health reserve]], and the most important kind here is [[ibnr-reserve|IBNR]] — *incurred but not reported*. Reserves are not a slush fund or spare profit; they are the bookkeeping recognition of money you already owe, just as binding as a bill on the desk.
It helps to separate the buckets. *Claim reserves* cover claims that have already happened: a known, reported-but-unpaid pile (close to case reserves), plus the unseen IBNR pile that experience says is surely out there. *Contract reserves*, by contrast, look forward within a long-dated policy — they are why a level-premium product whose true cost rises with age must hold money in the cheap early years to pre-fund the expensive later ones, the same pre-funding logic behind the policy reserves of the life rung. Short annual medical cover needs little contract reserve; long-term disability or long-term-care cover, which can pay for decades, needs a great deal.
So how do you estimate claims nobody has reported yet? The classic answer is the [[loss-development-triangle|loss-development triangle]]. Group claims by the month they were *incurred*, then watch how much had been *paid* one month later, two months later, three. A stable pattern emerges — say each incurred month is only 60% paid after one month, 85% after two, 95% after three. Flip the pattern around: if November had $600,000 paid by year-end and you know one-month-old claims are typically only 60% paid, then the *ultimate* is 600,000 / 0.60 = $1,000,000, so the remaining $400,000 is your IBNR for November alone. The newest months are the least mature, so they carry the most IBNR — and the most estimation risk. It is pattern-matching, not magic: you are using how old claims matured to forecast how young ones will.
Continuance tables: how long does a claim last?
For disability and long-term-care claims, IBNR is only half the worry. Once someone is *on claim*, the liability is the stream of monthly payments that will run until they recover or die. To reserve that, you need to know how long a typical claim *continues* — which is exactly what a [[morbidity-and-continuance-tables|continuance table]] records. It is the morbidity cousin of the life table you mastered on the survival rung: instead of "of 100,000 lives, how many are still alive at each age," it asks "of 1,000 claimants who all became disabled today, how many are still disabled after 3 months, 6, 12, 24?" The rate at which they leave is the claim-termination rate, which bundles two exits at once — recovery and death.
Continuance tables carry one sharp, counter-intuitive lesson: termination rates are highest early and then collapse. Most short-term illnesses resolve within weeks, so a flood of claimants recover quickly in the first months. But the longer someone has been disabled, the *less* likely they are to ever leave — a claimant still out at two years is far stickier than a fresh one. This is why the dangerous claims are the long ones: a small number of claimants who never terminate can dominate the reserve. The claim reserve for someone on disability is just the monthly benefit times the expected number of future months they will keep claiming, read straight off the continuance curve and lightly discounted.
Putting the machine together
Step back and see the whole loop. You *rate* by projecting next year's frequency × severity forward by trend, loading for expenses and margin, and deciding — within the law — how finely to slice the pool between community and experience, blended by credibility and defended against antiselection. You *reserve* by measuring what you already owe: IBNR for medical claims still in the mail, and continuance-based claim reserves for disability and long-term care still running. Then the loop closes: this year's actual claims, captured through reserves, become the experience that feeds next year's rate.
Hold onto the honest caveats, because they are where real practitioners earn their keep. A premium is only as good as next year's trend guess, and that guess is fragile. A community rate is fair by one definition and a magnet for antiselection by another. An IBNR is a model leaning on a development pattern that history may not repeat. A continuance reserve hides its true danger in the long, sticky tail of claims that never terminate. None of these is a flaw to be hidden — they are the very judgement that makes health actuarial work a profession rather than a spreadsheet. Master the loop, then stay humble about every number flowing through it.