Survival Models and Life Tables Long Term Actuarial Mathematics (I) (LTAM1) Don Hong Actuarial Science Program Middle Tennessee State University don.hong@mtsu.edu ©Spring 2022 Don Hong LTAM2 Survival Models and Life Tables Survival Models and Life Tables 1. Basic Probability Review 2. Cumulative, survival, and hazard rate functions 3. Future life time random variable Tx 4. Curtate Future Lifetime Random Variable Kx 5. Life Table Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table Probability Space and Random Variable I Definition 1. Given a set ⌦, a probability P on ⌦ is a function defined in the collection of all (subsets) events of ⌦ such that (i) P(;) = 0. (ii) P(⌦) = 1. (iii) If An , n = 1, 2, · · · , are disjoint events, then P 1 P([1 n=1 An ) = n=1 P(An ). ⌦ is called the sample space. I Definition 2. A random variable X is a function from the sample space ⌦ into R. We will abbreviate random variable into r.v. I For any r.v. X , we usually need to find its mean value for center point measurement and its variance or standard deviation for data distribution spread information. Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I Many insurance concepts depend on accurate estimation of the life span of a person. It is of interest to study the distribution of lives’ lifespan. The life span of a person (or any alive entity) can be modeled as a positive (r.v.) random variable. To model the lifespan of a life, we use age-at-death random variable X . The probability P(X > x) is called the survival function of the newborn. I For inanimate objects, age-at-failure is the age of an object at the end of termination. Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table 2. Cumulative, survival, and hazard rate functions I Definition 3. The cumulative distribution function of a r.v. X is FX (x) = P(X x), x 2 R. If Let X denote a newborn’s age, then corresponding to this random variable, the function S(x) = 1 FX (x) = P(X > x) represents a survival probability, therefore, it is called the survival function. I The survival function satisfies the following (i) S(x) is non-increasing, i.e. for each x1 x2 , S(x1 ) S(x2 ). (ii) S(x) is continuous from left, i.e. limh!0 S(x + h) = S(x). (iii) For each x 0, S(x) = 1. (iv) limx!1 S(x) = 0. 0 S (x) I The hazard rate function hX (x) = 1 fXF(x)(x) = S(x) , here we assume X S(x) is differentiable. It is also called the force of mortality in survival analysis, and denoted it as µ(x). Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I Relationship Formulas: 0 (x). (1) FX (x) + s(x) = 1, fX (x) = s Rx s0 (x) (2) µ(x) = s(x) , s(x) = e 0 µ(s)ds . I For any random variable X , its mean value E(X ) and variance VR(X ) are defined as follows. 1 • E(X ) = 1 xf (x)dx R1 2 • Var (X ) = E[(X µ) ] = 1 (x µ)2 f (x)dx = E(X 2 ) [E(X )]2 . • For new born’s haveR R 1 life time r.v. X , we 1 E(X ) =e̊0 = 0 s(x)dx and E(X 2 ) = 2 0 xs(x)dx. Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table Future life time random variable Tx I We are interested in analyzing and describing the future lifetime of an individual. Let (x) denote a life age x for x 0. Then the random variable describing the future lifetime, or time until death, for (x) is denoted by Tx , Tx 0. I The CDF of Tx is the probability that (x) dies in t years: Fx (t) = P[Tx t] = P[x < X x + t|X > x] = S(x) S(x + t) . S(x) We also denote this probability by using the notation t qx . I Then, the probability that (x) survives over t more years is R x+t S(x+t) µ(s)ds x =e t px = 1 t qx = S(x) . We can have t px = e I For very small x, S(x) S(x+ x) S(x) S 0 (x) x S(x) Rt 0 µ(x+s)ds . Pr [x < X x + x|X > x] = ⇡ = µ(x) x. Therefore, µ(x) x can be interpreted as the probability that a newborn who has attained age x dies between x and x + x. Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. I Relationship Formulas: • t px +t qx = 1, t px = s(x+t) , t qx = 1 s(x) Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table t px = s(x) s(x+t) . s(x) • The probability that (x) survives over t more years but cannot make S(x+t+u) over the next u years is t|u qx = S(x+t) S(x) = t px · u qx+t . • s+t px = s px · t px+s , • Denote 1 px = px and 1 qx = qx . For positive integer n, n px = px · px+1 · · · px+n 1 , where px+k = 1 px+k . P Pn 1 • Denote k| qx = k|1 qx . Then, 1 q = 1, and k=0 k| x k=0 Don Hong LTAM2 k | qx = n qx . 1. 2. 3. 4. 5. Survival Models and Life Tables Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I The pdf of Tx is d fx (t) = Fx (t) = dt S 0 (x + t) = S(x) S 0 (x + t) S(x + t) · = µ(x + t) · t px . S(x + t) S(x) I The expected value of Tx , or the mean value of T (x), denoted by e̊x , also called theRcomplete life expectancy of (x), is R1 R1 1 e̊x = E[Tx ] = 0 t · fx (t)dt = 0 t · t px µ(x + t)dt = 0 t px dt. (Ex. Show this by using Integrating by Parts) R R R I E[T 2 ] = 1 t 2 fT (t)dt = 1 t 2 · t px µ(x + t)dt = 1 2t · t px dt and thus, 0 2 Var [T ] = E[T ] 0 2 2 (E[T ]) = E(T ) Don Hong e̊2x . LTAM2 0 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I Example-1. Suppose that the survival function of X is s(x) = e x (x + 1), x 0. Find E[min(X , 10)]. I Solution: fX (x) = S 0 (x) = xe x . E[min(X , 10)] = = Z 1 0 min(x, 10)f (x)dx = x2 = ( 2)e ( + x + 1)|10 0 2 This value is also equal to e̊0:10| . x Z 10 0 xe 10e x x xdx + Z 1 10 10e (x + 1)|1 10 = 2 x xdx 12e 10 . I In general, for Tx , we have the term expectation of life: Rn R1 Rn E[min(Tx , n)] =e̊x:n| = 0 t ·t px µ(x +t)dt + n n·t px µ(x +t)dt = 0 t px dt. I Relationship formula: e̊x:m+n| =e̊x:m| + m px e̊x+m:n| . Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table Some commonly used forms for the force of mortality I It is sometimes convenient to describe the future lifetime random variable by explicitly modeling the force of mortality. Some of the parametric forms that have been used are: I Gompertz: µ(x) = Bc x , 0 < B < 1, c > 0 I Makeham: µ(x) = A + Bc x , 0 < B < 1, c > 0 I de Moivre’s: µ(x) = ! 1 x , 0 x < ! I Exponential: µx = , for 0 x 1. I Weibull: µx = kx n , for x > 0, k > 0, n > 1. Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I Example 2. [SOA LTAM Spring 2019 Q3] You are given: A life table uses a Makeham’s mortality model with parameters A = 0.00022, B = 2.7 ⇥ 10 6 , c = 1.124. 10 p50 = 0.9803. Calculate d ( q ) at t = 10. dt t 50 I Solution. At t = 10, 60 d ( q ) = f (10) = p µ(60) = (0.9803) · (A + B · c ) = 0.003158. t 50 50 10 50 dt Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table Curtate Future Lifetime Random Variable I We are often interested in the integral number of years lived in the future by an individual. This discrete random variable is called the curtate future lifetime and is denoted by Kx = bTx c. I We can consider the pmf (probability mass function) of Kx : Pr [Kx = k ] = Pr [k Tx < k + 1] = k| qx = k px qx+k . I We can find the mean value of Kx , also called the curtate expectation of life, denoted by ex , E[Kx ] = ex = 1 X k · k px qx+k = 1 X k · k px k =0 and the 2nd moment and the variance of Kx : E[K 2 ] = 1 X k =0 k 2 · k px qx+k = 2 k=1 Don Hong LTAM2 1 X k px . k =1 ex , V [Kx ] = E[K 2 ] (ex )2 . 1. 2. 3. 4. 5. Survival Models and Life Tables I Example 3. Suppose you are given: ( 0.04, µ= 0.05, Calculate e̊25:25| . I Solution. t p25 = e Rt 0 µ(25+t)dt = ( e e 0.04t Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table x < 40 x 40. 0 t < 15, . 15 t 25. , 0.04(15) · e 0.05(t 15) , R 25 R 15 0.04t R 25 0.04(15) e̊25:25| = 0 t p25 dt = 0 e dt + 15 e ·e 0.05(t 15) dt ⇡ 15.60. I Example 4. [SOA MLC S2016 Q2] You are given the survival function: S0 (x) = (1 x 1/3 ) , for 0 x 60. 60 Calculate 1000µ35 . 0 (x) I Solution. µx = SS(x) = 3(601 x) and thus, 1000µ35 = 13.3333. I In general, the so-called Generalized De Moivre’s Law is expressed as S0 (x) = (1 !x )↵ , for 0 x !. Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table 5. The Life Table I A life table is a tabular presentation of the mortality evolution of a cohort group of lives. I Beginning with `0 number of lives (e.g. 100,000), also called the radix of the life table, the (Expected) number of lives who are age x is `x = `0 · S0 (x) = `0 · x p0 . I The (Expected) number of deaths between ages x and x + 1 is denoted as dx = `x `x+1 and the (Expected) number of deaths between ages x and x + n is n dx = `x `x+n . I Conditional on survival to age x, the probability of dying within n years is n qx = n`dxx = (`x ``xx+n ) . I Conditional on survival to age x, the probability of living to reach age x + n is n px = 1 n qx = `x+n . `x Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I Examples of Life Table: • Illustrative Life Table (ILT) • Standard-Ultimate-Life-Table (SULT) • U.S. Life Table for the total population, 2004, Center for Disease Control and Prevention (CDC): x `x dx qx px e̊x 0 100,000 680 0.006799 0.993201 77.84 1 99,320 48 0.000483 0.999517 77.37 2 99,272 29 0.000297 0.999703 76.41 3 99,243 22 0.000224 0.999776 75.43 . . . . . . . . . . . . . . . . . . 50 93,735 413 0.004404 0.995596 30.87 51 93,323 443 0.004750 0.995250 30.01 52 92,879 475 0.005113 0.994887 29.15 53 92,404 507 0.005488 0.994512 28.30 . . . . . . . . . . . . . . . . . . 97 5,926 1,370 0.231201 0.768799 3.15 98 4,556 1,133 0.248600 0.751400 2.95 99 3,423 913 0.266786 0.733214 2.76 I For more life tables, you can browse CDC Life Tables Don Hong LTAM2 1. 2. 3. 4. 5. Survival Models and Life Tables Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I From a life table, we can have the following relationship formulas: P1 • `x = k =0 dx+k : the number of survivors at age x is equal to the number of deaths in each year of age for all the remaining years. Pn 1 • n dx = `x `x+n = k=0 dx+k : the number of deaths within n years is equal to the number of deaths in each year of age for the next n years. • the probability that (x) survives the next m years but dies the following n years after that can be derived using: m|n qx = m px n+m px Don Hong = n dx+m `x LTAM2 = `x+m `x+n+m `x . 1. 2. 3. 4. 5. Survival Models and Life Tables Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I From the relationship formula: Rx `x = `0 · S(x) = `0 · e 0 µ(s)ds , we can show that the force of mortality can be expressed in terms of life table function `0x as: µx = `x and with a simple change of variable, it is easy to see that µx+t = 1 `x+t • It follows immediately that: d `x+t dt = 1 t px · d dt (t px ) = t px µx (t). · d t px dt . I the curtate expectation of life can be expressed now as P1 P1 `x+k E[Kx ] = ex = k =1 k px = k=1 `x . I The n-year temporary curtate expectation of life is Pn Pn `x+k ex:n| = k=1 k px = k=1 `x . Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I Example 5. You are given the following extract from a life table: x `x 94 16,208 95 10,902 96 7,212 97 4,637 98 2,893 99 1,747 100 0 Calculate (i) 2| q95 . (ii) e95 . (iii) the variance of K95 , the curtate future lifetime of (95). (iv) e95:3| . I Solution. (i) 2| q95 = `97 `98 = 4637 2893 = 0.15997065. `95 10902 P1 P4 `95+k 7212+4637+2893+1747 ⇡ 1.5125. (ii) e95 = p = 95 k k =1 k =1 `95 = 10902 P4 2 2 2 2 d95+k (iii) Var (K95 ) = E(K95 ) e95 ⇡ 2.0984 since E(K95 )= k · ` = 4.386076. k =1 95 P3 `95+k (iv) e95:3| = = 1.3522. k=1 ` 95 Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I Example-6. [2020F SOA LTAM Writing Problem Q2] Consider the following expression for the force of mortality: µx+t = ↵e (x+t) . t (i) Show that for ↵ > 0, > 0, t px = exp[ ↵( e 1 )]. (ii) Derive an expression for t px when = 0. Don Hong LTAM2 Survival Models and Life Tables 1. 2. 3. 4. 5. Basic Probability Review Cumulative, survival, and hazard rate functions Future life time random variable Tx Curtate Future Lifetime Random Variable Kx Life Table I Example-7. [Written Q4. LTAM 2021F] (a) Show that ex = ex:n| + x pn ex+n . (b) You are given: (i) Mortality follows the Standard Ultimate Life Table (SULT). (ii)e87 = 6.56. Show that e90 = 5.2 to the nearest 0.1. Calculate the value to the nearest 0.01. (c) Let H = min(3, K87 ) denote the 3-year temporary curtate future lifetime of (87). Calculate the standard deviation of H. Don Hong LTAM2