Lesson 5 Survival Distributions: Mortality Laws Reading: Actuarial Mathematics for Life Contingent Risks 2° edition 2.2, 2.3, 2.7, 3.4 ‘There are two approaches to defining So(x), the survival function for the age at death. One is to build a life table defining the probability of survival to each integral age, and then to interpolate between integral ages. We shall learn how to interpolate in Lesson 8. The other approach is to define Sp(x) as a continuous function with parameters. Usinga parametric function has the advantage of capturing the distribution with only a small number of parameters, which makes the table more portable. The function may have nice properties which simplify computations. Such parametric functions are called mortality laws. This lesson will discuss two types of mortality laws. ‘The first type of mortality law is a parametric distribution that reasonably fits human mortality or some other type of failure over a wide range of ages. However, itis hard to write a deep exam question using such a function, and they rarely appear on exams. ‘The second type of mortality law is a simple parametric distribution that is quite unrealistic for human mortality, but is easy to work with. Two of the morlality laws we discuss in this section are virtually the only ‘ones that allow easy computation of insurances and annuities in closed form without numerical methods. They appear frequently on pre-2012 exams and in the SOA sample questions. However, they appear only occasionally on current exams. 5.1 Mortality laws that may be used for human mortality ‘To get some idea of what the survival curve looks like, we will look at graphs. These graphs are imitations of the graphs in Actuarial Mathematics for Life Contingent Risks. However, they are based on US. 2013 Life Tables, ‘whereas that textbook uses three otier tables. The U.S. 2013 Life Tables are cut off at age 100; there is very little data above that age Figure 5.1 shows the survival function for females at three starting ages: 0, 40, and 80. These functions are related; for example S4o() = So(40 + £)/S0(40). So each curve is a truncated, shifted, and scaled version of the previous one, S,(4) is the same as :px, and if x is fixed, then Try; = ly sp #8 a constant multiple of ip, s0 the curves for ip and Irs; look the same as the curves for S(t) Figure 52 shows the probability density function for females at three starting ages: 0, 40, 80. In this graph and the following ones, I used simple approximations for fiz. The three curves are related in the same way as the three $:(t) curves are related: truncate and scale to go from one curve to another. We see that the most likely age at death is approximately 85. If we had drawn a curve for fs({), x > 85, it would be monotonically decreasing. Since fy(t) = jPy fxs, CUEVES for ip jlxae and ly 1x Would look the same. ‘The most interesting graph is the one for force of mortality. Figure 5.3 graphs male and female forces of mortality. Buta clearer way to see the pattern is to use a logarithmic scale, as in Figure 5.4. Characteristics of the In jr curve, based on this graph, are 1, Mortality is higher for males than for females at all ages, except at around age 10 male mortality is, (strangely) below female mortality. The lines appear to merge as age increases, and some believe male mortality may be lower than female at very high ages. 2. Mortality is very high at birth but quickly drops until about age 10. 3. For the male table only, there is a hump in the 20s. For both sexes, there is a rapid increase in mortality in the late teens. Both of these are caused by high accident rates, for example from youth driving, SOA LTAM Stuy Manual—1" tion 2 printing 65 Copyright S208 ASNT SEBO woiswantcom | FUDGE wraulwaty.com66 5. SURVIVAL DISTRIBUTIONS: MORTALITY LAWS Poo NN ° 20 w w w T00 Figure 5.1: Survival function for three ages o = fa) 0.09 = fait oe = halt) 07] 0.06 0.05} oy 0.05} .0n| 001 °% 20 40 60 80 100 ' Figure 5.2: Probability density function for three ages SOA LTAM Stuy Manual—1" tion 2 printing Copyright S208 ASNT SEBO woiswantcom | FUDGE wraulwaty.com5.1. MORTALITY LAWS THAT MAY BE USED FOR HUMAN MORTALITY o7 Be oa! = Female = Male 03 02 on 0 20 0 60 30 100 Figure 5.3: Force of mortality for males and females = Female = Male o.o1 001 o 00015 20 40 60. 80, 100 * Figure 5.4: Force of mortality for males and females with logarithmic scale SOA LTAM Stuy Manual—1" tion 2 printing Copyright S208 ASNT SEBO woiswantcom | FUDGE wraulwaty.com68 5. SURVIVAL DISTRIBUTIONS: MORTALITY LAWS 4, Most importantly, the graphs are virtually linear starting at about age 40. ‘The last characteristic leads us to our first mortality law. 5.1.1 Gompertz’s law Ifwe assume In jis a straight line, we can solve for the parameters (by linear regression or some other means): Inpix = a+ Bx Exponentiating, we get Gompertz’s law: bs = Be* Ga) with appropriate parameters B and ¢ > 1.1 Aswe saw from Figure 5.4, this law fitsages 40and above fairly well. Gompertz’s law isa good approximation to mortality for ages past the accident hump. ‘The survival function under this law is ese f'n) woo 22=9) 62 Since the law has two parameters, you can solve forall functions if you are given two values of mortality. Examrte SA Mortality fora life age 20 follows Gompertz’s law. You are given j1s9 = 0.001 and igo = 0.15. Determine sqpav. Answer: Set up two equations for ages 30 and 89. InB +30Ine = In0.001 InB+80Ine = In0.15, 50Inc = In 150 cellm3050 1.105406 In 150) 50 9.914136 tn = ino 9 8 = o.0000494704 Using formula (5.2), 0,0000494704(1.105406"" )(1.105406™ ~ 1) sp = xp (SRN ee =D) - a] Tn 1.105406 5.1.2. Makeham’s law Makeham improved Gompertz’s law by adding a third parameter A: He = A+ Bet 63) “fe = dhs exponential moray fe <1, then the nepal eds dows nat approach infin a + SOA LTAM Stuy Manual—1" tion 2 printing Copyright S208 ASNT SEBO woiswantcom | FUDGE wraulwaty.com5.1. MORTALITY LAWS THAT MAY BE USED FOR HUMAN MORTALITY 69 A represents the constant part of the force of mortality, mortality that is independent of age and is due to accidental causes. Be*, with ¢ > 1, represents mortality resulting from deterioration and degeneration, which increases exponentially by age. Makeham’s law provides a good fit for ages above 20. ‘As we know, adding, A to 1: multiplies the survival function by e~*", so building on equation (5.2), _ Bet(e' - 1) ina Ga) moon (-at ‘This formula is one of the five on the formula sheet you get at the exam. Since the law has three parameters, you can solve forall functions if you are given three values of mortality. However, numerical methods may be needed to solve for the parameters. Examete SB Mortality follows Makeham’s law. You ate given j110 = 0.0014, j= 0.0024, and jtso = 0.0042. Determine sqpav. Answer: Write out the three equations for the three y's AtBe= A+B A+Be® Bele! 1) = Bc (cl0 B A + 0.000684444(1.8) = 0.0014 A= 0.00015 ‘Therefore, the probability of surviving 50 years is (In the following expression, we use cl = 1.8, so c2? = (c)? = 1.8% and c® = (cl0)F = 1.85) _ 0,000694444(1.8°)(1.8° - 1) spn = exp (-0.001560) - SADE 2) — aa] o Actuarial Mathematics for Life Contingent Risks mentions a generalization of Makeham’s law: GM(r,s) (GM standing for Gompertz-Makeham) having the form fix = h(x) + exp(h2(x)) where it! and fi? are polynomials of degrees r and 5 respectively. Makeham’s law is GM(0, 1) and Gompertz’s law adds the further constraint that i; = 0. Survival probabilities usually cannot be computed in closed form for these jtx’s, s0 I doubt they will appear on an exam, 5.1.3. Weibull Distribution Rarely an exam question will use the Weibull distribution, although it won't be called that. The exam question ‘would just say jy = kx" using some specific k and For the Weibull distribution, the cumulative distribution function is /0y F(x) =1- SOA LTAM Stuy Manual—1" tion 2 printing Copyright S208 ASNT SEBO woiswantcom | FUDGE wraulwaty.com