Survival Models: SOA FAM
A survival model describes the time until an event, often future lifetime.
- For a life table with lₓ=1,000, lₓ₊₁=980 and lₓ₊₂=950, one-year survival is 0.98 and two-year survival is 0.95.
- Conditional survival from age x+1 to x+2 is 950/980, a different probability.
On this page12 sections
- Start with the survival function
- Complete and curtate future lifetime
- Force of mortality
- Fractional ages and select tables
- Original life-table calculation
- Derive interval probabilities from the life table
- Survival functions and future lifetime
- Complete and curtate future lifetime are distinct
- Fractional ages require an assumption
- Use mortality models with the right population basis
- Connect mortality to a payment model
- Check the life table arithmetic
Start with the survival function
Let Tₓ denote future lifetime for a life aged x. The survival function is the probability that future lifetime exceeds t, often written as t pₓ. The corresponding probability of death within t years is t qₓ=1−t pₓ. If a life table records lₓ survivors at age x, then for integer t, t pₓ=lₓ₊ₜ/lₓ, assuming the table’s cohort interpretation.
With lₓ=1,000, lₓ₊₁=980 and lₓ₊₂=950, one-year survival is 0.98 and two-year survival 0.95. The second-year conditional survival is 950/980≈0.9694. The year-two death probability is 30/980 conditional on reaching x+1, but 30/1,000 as a fraction of the starting cohort.
Complete and curtate future lifetime
Complete future lifetime Tₓ can take any nonnegative real value. Curtate future lifetime Kₓ=floor(Tₓ) counts the number of complete years lived. They answer different questions. A continuous-time benefit at death may depend on Tₓ, while an annual end-of-year benefit often depends on Kₓ and the integer year of death.
If the life dies between ages x+2 and x+3, then Kₓ=2 while Tₓ is between 2 and 3. The probability mass for Kₓ=k is k|qₓ, the probability of survival k years and death during the next year. The distribution of Tₓ instead uses the survival function between integer ages and an assumed fractional-age model or parametric model.
Force of mortality
For a continuous survival model with survival function Sₓ(t), the force of mortality at future duration t is μₓ₊ₜ=−d/dt ln Sₓ(t), where the derivative exists. It is an instantaneous rate conditional on surviving to t, not the same as the probability of death during an entire year. Integrating the force over an interval gives the cumulative hazard, which determines survival over that interval.
Under a constant force μ over one year, survival is e^(−μ) and death probability is 1−e^(−μ). For small μ these are close to familiar rate approximations, but they are not identical. State the model before using the shortcut. The syllabus expects calculating survival, mortality and force for parametric models and life tables.
Fractional ages and select tables
Life-table data often appears at integer ages, while a problem may ask about fractional duration. The October syllabus explicitly includes appropriate fractional-age assumptions where necessary. Identify the assumption supplied by the problem or syllabus context before interpolating. Do not silently assume uniform distribution of deaths when another assumption is specified.
Select tables distinguish mortality for lives recently selected from ultimate rates after a selection period. A select notation can depend on both issue age and duration since selection. Read subscripts and superscripts carefully. The learning outcome includes identifying and applying select life tables, so practice the notation rather than treating all q values as ultimate rates.
Original life-table calculation
Suppose lₓ=1,000, lₓ₊₁=980, lₓ₊₂=950. Then qₓ=(1,000−980)/1,000=0.02. One-year survival pₓ=0.98. The probability of death during the second year conditional on reaching age x+1 is qₓ₊₁=(980−950)/980≈0.03061. The joint probability of surviving year one and dying in year two is 0.98×0.03061=0.03.
The distinction between a conditional rate and an initial-cohort probability is easy to miss. If a prompt asks “of those alive at x+1,” use 980 in the denominator. If it asks “what fraction of the original 1,000 die in the second year,” use 1,000. Write the population at risk before substituting numbers.
Derive interval probabilities from the life table
A life table supplies counts or probabilities tied to age. If l_x lives reach exact age x, then d_x=l_x-l_{x+1} and q_x=d_x/l_x, while p_x=l_{x+1}/l_x. These are one-year conditional probabilities for a life aged x, and p_x+q_x=1. For multiple years, multiply conditional survival probabilities across attained ages; do not treat the original cohort count as the denominator at every age.
For an illustrative table with l_40=1,000, l_41=980 and l_42=950, p_40=.98 and the two-year survival probability from 40 to 42 is.95. Conditional survival in the second year given survival to 41 is 950/980, about.9694. The unconditional probability of dying in the second year from age 40 is 30/1,000=.03, while the conditional probability given survival to 41 is 30/980. The question’s conditioning language determines which applies.
Survival functions and future lifetime
For a continuous future lifetime T_x, the survival function is often written _tp_x=P(T_x>t). It describes surviving t years from age x. The distribution function is 1-_tp_x, and a density, when it exists, is the derivative of the distribution function. Use the source’s notation carefully because p_x without a time subscript often denotes one-year survival, while _tp_x denotes survival for t years.
The hazard or force of mortality at duration t is the instantaneous death rate conditional on survival to t. If it is constant at mu, then _tp_x=e^(-mu t) and the density is mu e^(-mu t). With a changing force, survival is the exponential of the negative integrated force. A constant-force assumption should come from the problem or model, not from the fact that it makes the algebra convenient.
Complete and curtate future lifetime are distinct
Complete future lifetime measures the exact remaining lifetime T_x, while curtate future lifetime K_x counts the number of complete years lived after age x. For an integer k, K_x=k means survival through k full years and death before the next birthday. Therefore its probability is _kp_x q_{x+k} under the standard discrete notation. Complete lifetime supports continuous payment timing; curtate lifetime is natural for benefits paid at year-end.
This distinction changes the present-value random variable. A benefit paid immediately at death depends on the exact death time and can involve a continuous discount factor. A benefit paid at the end of the year of death uses the integer year of death. Substituting the curtate model into a continuous-payment question changes timing, even if the mortality table is the same.
Fractional ages require an assumption
A life table commonly reports integer-age values, while a question can ask for survival over part of a year. Fractional-age interpolation requires an assumption about mortality between birthdays, such as uniform distribution of deaths within the year or constant force of mortality. These assumptions produce different fractional probabilities. Use the assumption stated in the problem or the syllabus source; do not silently choose one.
Under a uniform-distribution-of-deaths assumption, survival over a fraction t of a year can be interpolated linearly from the number alive at the endpoints. Under constant force, the fractional survival probability is a power of the one-year survival probability. They are close for small mortality but not identical. A careful solution labels the interpolation basis before calculating.
Use mortality models with the right population basis
A life table summarizes mortality for a defined basis, age, period or insured population. A value derived from it is conditional on that modeled basis. Do not assume a table automatically includes selection, underwriting, future improvement or a particular extra risk. The syllabus includes specified extra-risk methods; an adjustment should follow the method named in the problem.
When comparing two ages or durations, keep the cohort basis consistent. A period table and a cohort projection may represent different mortality assumptions. If the problem supplies l_x or q_x directly, use those inputs rather than importing outside mortality rates. The core skill is translating a table into the probability needed for the cash flow.
Connect mortality to a payment model
Mortality probabilities become financially meaningful when attached to a cash flow. For an end-of-year death benefit, the probability of death in year k weights the payment at time k. For an annuity, survival to the payment date weights the cash flow. Both use the same underlying life table, but they ask about different events.
Write each event before summing. A benefit paid at the end of year two of death uses survival through year one and death during year two. An annuity due at time one uses survival to that payment date. This event-first method reduces indexing mistakes.
Check the life table arithmetic
Life table counts should not increase with age under the usual cohort interpretation. The difference l_x-l_(x+1) gives deaths during the interval, and dividing by l_x gives q_x. The surviving count ratio gives p_x. Confirm p_x+q_x=1 and that multi-year survival does not exceed one.
If a problem gives probabilities rather than counts, multiply conditional survival probabilities across periods. Do not multiply unconditional probabilities that use different starting populations without first checking their definitions. A stated table basis controls the interpretation.
Common questions
What is the difference between tpx and tqx?
Tpx is survival for t years; tqx is death within t years, under standard notation.
How do I use lx values?
For integer t, divide l at age x+t by l at age x for cohort survival.
What is curtate future lifetime?
It counts complete future years lived, unlike continuous complete future lifetime.