JOVANA
Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

Survival Models: The Future-Lifetime Random Variable

Everything in life actuarial work rests on one quiet idea: how much longer a life will last is a random variable. Meet that variable, and meet the survival function — the curve that tells you the chance of still being here at any future age.

From interest to lives: a new kind of uncertainty

In the Interest rung you learned to move money cleanly through time: accumulate a present amount forward, discount a future one back, and balance everything in an equation of value. But every one of those calculations quietly assumed the payment *would* happen on its date. Life insurance and pensions break that assumption. A death benefit is paid only *if* and *when* someone dies; a pension cheque arrives each month only *while* the retiree is still alive. The timing is not on a calendar — it is in nature's hands, and it is uncertain. This rung gives that uncertainty a precise shape.

You already have the tool for it. In the Probability rung you met the [[act-random-variable|random variable]] — a number whose value is decided by chance, summarised by a distribution. The whole trick of life actuarial work is to recognise that *time of death* is exactly such a number. We cannot know when any one person will die. But across a large pool of similar lives, the pattern of deaths over the years is remarkably stable and measurable — and a random variable is precisely how we capture "unknown for the individual, predictable in aggregate". Hold that thought: it is the same risk-pooling logic from the Foundations rung, now aimed at the one event insurance cares about most.

Two clocks: age at death and remaining lifetime

There are two natural ways to time a death, and life actuaries use both. The first is the [[age-at-death-random-variable|age at death]], written X: the total length of a newborn's life, measured from birth. It is a continuous random variable that can take any value from zero upward, and its distribution describes how a whole cohort of newborns will eventually be spread across the ages at which they die. X answers the question "how long does a fresh life last, start to finish?"

But an insurer rarely deals with newborns. It deals with a 45-year-old buying a policy, or a 67-year-old starting a pension — people who have *already survived* to some age x. What matters then is not the whole life, but what remains of it. That is the [[future-lifetime-random-variable|future-lifetime random variable]], written T(x): the time still to be lived by someone known to be alive at exact age x. If a person aged x dies at age X, then T(x) = X − x. T(x) is the true protagonist of this rung — almost every life insurance and annuity value is, underneath, an expectation taken over T(x).

The survival function: the chance of still being here

Now we describe T(x) the way the Probability rung taught us to describe any random variable — but with one telling choice. For most variables we reach first for the [[cumulative-distribution-function|cumulative distribution function]], the probability of being *at or below* a value. For lifetimes we flip it over and lead with its complement, because the question that pays the bills is "will the life *last*?" rather than "will it *fail* by now?" That flipped quantity is the [[survival-function|survival function]], written S(x): the probability that a newborn survives *past* age x. By definition S(x) = 1 − F(x), where F is the ordinary distribution function of age at death.

Picture S(x) as a curve. At birth S(0) = 1 — everyone is, by definition, alive at the very start. As age rises the curve can only slide downward or stay level; it never climbs, because once you have failed to survive past some age you cannot un-die. Far out at the oldest ages it sinks toward zero. The *steepness* of the slide at any age tells you how thickly deaths are clustering there: where the curve plunges, lives are being lost fast; where it merely eases down, the cohort is dying slowly. So a single monotone curve carries the entire story of how a population dies — gently in midlife, then ever faster in old age.

Imagine 100,000 newborns and a survival function S(x):

   S(0)  = 1.0000   ->  100,000 alive at birth
   S(40) = 0.95     ->   95,000 still alive at exact age 40
   S(65) = 0.82     ->   82,000 still alive at exact age 65
   S(90) = 0.28     ->   28,000 still alive at exact age 90

P(a newborn dies between 65 and 90) = S(65) - S(90)
                                    = 0.82 - 0.28 = 0.54
S(x) read as the fraction of a starting cohort still alive at age x; the probability of death in any age band is just the drop in S across that band.

Survival, density, and the shape of death

The survival function and the distribution of death are two faces of one object, and it is worth holding both in view. F(x) = 1 − S(x) is the probability of dying *by* age x; its slope, the [[probability-mass-and-density-functions|probability density]] f(x), measures how concentrated deaths are right around age x. Because S = 1 − F, the density is exactly the *rate at which S is falling*: where survival drops fastest, the death density peaks. For a typical human population that peak sits in the high seventies or eighties — not because death is impossible earlier, but because that is where the thinning cohort and the rising per-person risk multiply together most heavily.

Be careful, though, not to confuse two different "risks". The density f(x) measures risk relative to the *whole original cohort*; it eventually shrinks at the oldest ages simply because almost no one is left to die. That does *not* mean a 100-year-old is safer than a 70-year-old. The per-person danger keeps climbing — there are just fewer people left to count. Separating those two ideas — deaths per starting newborn versus risk per survivor still standing — is precisely the job of the [[force-of-mortality|force of mortality]], and untangling the [[survival-density-hazard-relationship|relationship between survival, density, and hazard]] is the work of the next guide. For now, simply notice that a low death *count* at extreme ages hides a very high death *rate*.

Why the actuary cares: from a curve to a price

The survival function is not abstract decoration — it is the input that lets an insurer put a number on a promise. Suppose a 65-year-old buys a contract paying 1,000 *only if they survive one more year*. Using the cohort above, the conditional chance of surviving from 65 to 66 is the ratio of survivors: roughly S(66) ÷ S(65). Multiply that probability by the 1,000 to get the *expected* payment, then discount it back through the interest gears from the last rung. That two-step — a probability of payment times a discount factor — is the seed of the [[actuarial-present-value|actuarial present value]], the single idea on which the entire Life Contingencies rung is built.

The same curve also tells us how long, on average, a life will run — its [[life-expectancy|life expectancy]] is simply the expected value of T(x), the area swept out under the survival curve from age x onward. There is a subtle choice hiding here that the rung will sharpen: do we count whole completed years only, or the exact fractional time lived? That distinction between [[curtate-vs-complete-future-lifetime|curtate and complete future lifetime]] changes annuity and insurance values by a small but real amount, and a good actuary never blurs it. And in practice we rarely carry S(x) as a smooth formula at all — we tabulate it, age by age, into the [[life-table|life table]], the workhorse the next guides turn to again and again.