The risk that hides inside good news
Everything you have built so far in this rung — a life annuity that pays until death, a defined-benefit pension promising income for life — rests on one fragile number: *how long, on average, will the recipients live?* Pick a mortality table, run it through the actuarial present value machinery from earlier rungs, and out pops a price. The trouble is that the table is a forecast, and forecasts of human lifespan have been wrong in one direction for over a century. People keep living longer than the table said they would. That systematic, one-sided error is longevity risk.
It helps to split the danger in two. *Idiosyncratic* longevity risk is the chance that one particular annuitant outlives the average — Mrs. Tan draws her pension for 40 years while the table expected 20. This kind of risk is what pooling tames beautifully: across thousands of lives, the ones who die early subsidise the ones who live long, and the average behaves. But *systematic* longevity risk is different and far nastier. It is the chance that the *whole population* lives longer than assumed — that the average itself was wrong. No amount of pooling helps here, because every life in the pool drifts the same way at once. That second kind is the one that keeps pension actuaries awake.
How a few extra years rewrites the price
Let us make the cost concrete, the way the cost-method machinery from earlier in this rung would. Take a pension paying 1 unit a year for life to someone aged 65. Discounting future payments and weighting each by the chance of surviving to receive it gives the annuity value — the lump sum needed today to fund the promise. The headline figure that summarises it all is life expectancy at 65: how many more years, on average, the payments are expected to run.
Pension: 1 per year for life from age 65, interest i = 4% Assume life expectancy at 65 = 20 yrs -> annuity value ~ 13.6 If people actually live 21 yrs -> annuity value ~ 14.0 (+3%) If people actually live 23 yrs -> annuity value ~ 14.8 (+9%) Rule of thumb at these ages and rates: +1 year of life expectancy ~ +3% to +4% on the cost of the promise
Now feel why this is so dangerous. A 9% under-pricing on a *single* policy is survivable. But a pension fund holds tens of thousands of these promises, all priced off the *same* assumed table. If that table understated lifespans by three years, the whole liability is understated by nearly a tenth — billions, for a national scheme — and there is no offsetting good news anywhere in the book to absorb it. The error is correlated across every life at once. This is exactly why longevity is a valuation assumption that gets argued over so fiercely: a small, plausible-sounding tweak to the assumed mortality moves the reported liability by more than almost anything else.
Trends, not snapshots: pricing a moving target
The deeper fix is to stop treating mortality as a fixed photograph. A table built from deaths over 2015–2020 describes the *past*. A 65-year-old retiring today will mostly draw their pension in the 2040s and 2050s, by which time medicine and living standards will likely have nudged death rates lower still. So actuaries layer a mortality improvement projection on top of the base table: an assumed annual rate at which death rates fall in the future. A common shorthand is something like "mortality improves about 1.5% a year" — meaning each year's death rate is assumed to be a bit below the last.
Be honest about what this is, though: a projection is a *forecast of a forecast*, and the historical record warns us it has been too pessimistic more often than not. Past improvement has come in surges and pauses nobody predicted — the great twentieth-century declines in infant and cardiovascular mortality were not foreseen, and recent years have seen improvement unexpectedly *stall* in some countries. The honest posture is humility: pick a central improvement assumption, then stress-test it. What does the liability look like if improvement runs at 2.5% instead of 1.5%? That gap — not the central estimate — is the real shape of the longevity risk.
Who can carry it — and the tools to move it
Once you see longevity risk clearly, the great pension design debate reads differently. In a defined-contribution plan, the worker carries longevity risk personally: they retire with a pot, and if they live to 100 they may simply run out. In a defined-benefit plan, the *sponsor* carries it — the promise is for life, whatever life turns out to be. The slow global retreat from DB toward DC is, in large part, employers handing this exact risk back to individuals, who are the worst-placed of all to bear it. Which is also why the lifelong life annuity is so valuable in theory yet so rarely bought — the famous annuity puzzle from earlier in this rung.
When a DB sponsor wants longevity risk off its books, a market now exists to move it. Three tools matter. A *buy-in* or *buy-out* hands the liabilities to an insurer, who is paid a lump sum to take over the promises. A *longevity swap* is narrower: the pension keeps paying its members, but swaps its uncertain, lengthening cash flows for a fixed schedule with a bank or reinsurer — paying a premium to convert "however long they live" into "this much, for certain." And joint-and-survivor designs spread one more layer of longevity uncertainty across a couple's two lives. In every case the principle is the same: longevity risk does not vanish, it is *transferred* to whoever is largest, most diversified, and best able to hold it.
Social insurance: the same arithmetic, written for a nation
Zoom out to the largest pension promise of all: the state. Social insurance schemes — public pensions, retirement systems run by governments — face the very same longevity arithmetic, only at the scale of an entire population, and usually funded very differently. Most are pay-as-you-go: today's workers' contributions are not saved in a fund for themselves; they are paid out *immediately* as today's retirees' pensions. The implicit promise is that the next generation of workers will, in turn, pay for you. It works as long as there are enough of them.
And there is the demographic pinch, honestly stated. Two forces squeeze pay-as-you-go at once: people live longer (more years of pension per retiree) *and* birth rates have fallen (fewer workers behind each retiree). The crude gauge is the old-age dependency ratio — retirees per working-age adult. In many countries it has moved from roughly four or five workers per retiree toward two, and is still tightening. There is no clever financial instrument that makes this go away, because at the national level the risk cannot be transferred to anyone — there is no larger pool. A society can only respond with real levers: a later retirement age, higher contributions, lower benefits, more immigration, or higher productivity. Pretending otherwise is the one thing an honest actuary will not do.
Two honest caveats round this out. First, longevity is not improving equally for everyone — the rich and well-educated tend to gain more years than the poor, so a flat rise in retirement age quietly takes the most from those who live the least. A good social-insurance design has to reckon with that fairness, not just the average. Second, longevity is genuinely *good news*: the alternative to an ageing population is a dying one. The actuary's job is not to lament longer lives but to price them honestly and help society fund the promise — so that living to 100 is a triumph to celebrate, not a bill nobody planned to pay.
Putting it together: how to think about longevity
- Separate the two risks. Ask whether your worry is one person outliving the average (pooling handles it) or the whole population outliving the table (pooling does not). Only the second is true systematic longevity risk.
- Use a cohort view, not a snapshot. Build the liability on a base table plus an explicit mortality-improvement projection, and remember a projection is a forecast that has historically erred toward pessimism.
- Stress-test, do not just point-estimate. The central life expectancy is the start; the difference between a 1.5% and a 2.5% improvement assumption is where the real risk — and the capital you must hold — actually lives.
- Ask who is carrying it. In DC it is the individual; in DB the sponsor; in social insurance the next generation. Risk can be transferred via buy-outs and longevity swaps — but at the national level there is no larger pool, only real policy levers.
That is the whole arc of this rung in one idea. Interest theory told you how to value a stream of payments; survival models and life-contingent functions told you how to weight each payment by the chance of being alive to receive it; premiums and reserves told you how to charge and hold for it. Longevity risk is what remains when you admit that the survival probabilities themselves are uncertain — and bending in one direction. Carry that humility forward: a model is a disciplined guess, not the future, and the actuary's craft is to price honestly *and* hold enough margin for the guess to be wrong.