You are on page 1of 20

SURVIVAL MODELS

Ruhiyat

Department of Mathematics
IPB

Bogor, 2020

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 1 / 20


Let (x) denote a life exactly age x, x 0.
Let Tx represent the time until death of (x).
Tx is the future lifetime random variable of (x).
Tx is a continuous random variable.
The age at death would be x + Tx .

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 2 / 20


Let’s consider the case where x = 0 (newborn).
T0 represents the future lifetime random variable for (0).
The cdf:
F0 (x) = Pr (T0 x) .

The survival function:

S0 (x) = Pr (T0 > x) = 1 F0 (x) .

The pdf:

d d
f0 (x) = F0 (x) = [1 S0 (x)] = S00 (x) .
dx dx

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 3 / 20


T0 is a continuous random variable, so Pr (T0 = x) = 0, 8x.
The pdf, f0 (x), describes the likelihood of T0 being "near" x.

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 4 / 20


Let’s take a look at T0 and Tx .
If we have S0 (x) or F0 (x) or f0 (x) for all x, then we can derive the
distribution of the random variable Tx for any x. All we need to do is
condition on the event T0 > x.
Let’s consider the death of (x) at age x + Tx .

Pr (Tx t) = Pr (T0 t + xjT0 > x)


Pr (x < T0 t + x)
=
Pr (T0 > x)
F0 (x + t) F0 (x)
Fx ( t ) = .
1 F0 (x)

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 5 / 20


Sx (t) = Pr (Tx > t) = 1 Fx (t)
F0 (x + t) F0 (x)
= 1
1 F0 (x)
1 F0 (x + t)
=
1 F0 (x)
S0 (x + t)
= .
S0 (x)

Let’s rearrange our expression. We have:


S0 (x + t) = S0 (x) Sx (t) .

Furthermore, for any t, u > 0,


Sx (t + u) = Sx (t) Sx+t (u) .

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 6 / 20


Conditions on S0 (x) or Sx (t)
The …rst three conditions are required.
1 Sx (0) = 1.
2 lim Sx (t) = 0.
t! ∞
3 Sx (t) is a non-increasing function of t.
The next three are desirable.
1 Sx (t) is di¤erentiable.
2 lim t Sx (t) = 0.
t! ∞
3 lim t2 Sx (t) = 0.
t! ∞

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 7 / 20


The force of mortality: µx
µx is easier to estimate from data, and interpret.
It is also called a hazard rate or transition intensity.
The de…nition of µx :

Pr (T0 x + dxjT0 > x)


µx = lim .
dx!0 dx

µx represents the instantaneous rate of mortality acting on a person


currently age x.
µx is not a probability. However, µx dx Pr (T0 x + dxjT0 > x)
for small dx.

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 8 / 20


Some properties associated in the force of mortality:
f0 ( x )
1 µx = .
S0 (x)
Note: We rearrange this to get f0 (x) = S0 (x) µx .
fx ( t )
2 µ x+t = .
Sx (t)
Note: µx+t is sometimes written as µx (t). Again, we can rearrange
this to get fx (t) = Sx (t) µx+t .
d
3 µx = [log S0 (x)].
dx
Important note: log natural log = ln.
d
4 µ x+t = [log Sx (t)].
dt
Rt
5 Sx (t) = exp µ dr .
0 x+r

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 9 / 20


Exponential Model - Constant Force of Mortality (CFM)

λx
S0 ( x ) = e , x > 0; 0, otherwise.
λx
f0 (x) = λe , x > 0; 0, otherwise.

λx
) F0 (x) = 1 e .

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 10 / 20


T0 Exp(λ)

S0 (x + t) e λ (x+t) λt
) Sx ( t ) = = =e = S0 (t)
S0 (x) e λx
(Memoryless property!)

Also,

f0 ( x ) λe λx
) µx = = =λ
S0 (x) e λx
(Constant force of mortality)

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 11 / 20


De Moivre Model - Uniform Distribution of Deaths (UDD)

x ω x
S0 (x) = 1 = , x ω; 0, otherwise.
ω ω
1
f0 (x) = .
ω
T0 U (0, ω ) .

S0 (x + t)
) Sx (t) =
S0 (x)
x+t
1 ω t
= x =1 , 0 t ω x; 0, otherwise.
1 ω ω x

Note: ω represents the limiting age in the model.


Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 12 / 20
Tx U (0, ω x)
1
f0 ( x ) ω 1
) µx = = ω x =
S0 (x) ω
ω x
(increasing force of mortality)

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 13 / 20


Gompertz Model

µx = B cx

Z t
) Sx (t) = exp µx+r dr
0
Z t
B cx
= exp B cx+r dr = exp ct 1
0 log (c)

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 14 / 20


Actuarial notation for probability:

Sx (t) = Pr (Tx > t) = t px .


Fx (t) = Pr (Tx t ) = t qx .
t j u qx = Pr (t < Tx t + u) .

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 15 / 20


Some results with notation
1 t px + t qx = 1.
2 Convention
If t = 1 year, we often omit it.
For example, 1 px is written as px .
3 Survival model results
fx (t) = t px µx+t .
d
µ x+t = [log (t px )].
dt R
t
t px = exp 0 µx+r dr .
0 px = 1, lim t px = 0, and t px is a non-increasing function.
R t t! ∞
t qx = 0 u px µx+u du.

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 16 / 20


Mean of Tx
Z ∞
E ( Tx ) = t fx (t) dt
Z0 ∞
= t t px µx+t dt
0
Z ∞
= t px dt
0

Notation: E (Tx ) = ex
ex complete expectation of life (x).

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 17 / 20


Median
The median of Tx is a value of t50 such that Pr (Tx t50 ) = 0.5 or
Pr (Tx > t50 ) = 0.5.
Mode
Mode of Tx is the most likely value.
We need to maximize fx (t) = t px µx+t .
Variance of Tx
2 2
Var (Tx ) = E Tx ex = E Tx2 ex

where
Z ∞
E Tx2 = t2 fx (t) dt
0
Z ∞
= t2 t px µx+t dt
0
Z ∞
= 2 t t px dt.
0

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 18 / 20


Let Kx = bTx c.
Kx represents the number of whole years of future life for (x).
Kx is the curtate future lifetime random variable.
Kx is a discrete random variable de…ned over the range f0, 1, 2, . . .g.
Kx has a probability mass function (pmf)

Pr (Kx = k) = Pr (k Tx < k + 1) = k jqx


= k px k+1 px
= k+1 qx k qx
= ( k p x ) ( qx + k )

The cdf of Kx
k
Pr (Kx k) = ∑ Pr (Kx = i) = 1 k+1 px = k + 1 qx
i=0

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 19 / 20


Mean of Kx

E ( Kx ) = ∑ k k j qx
k =0

= ∑ k px .
k =1

Notation: E (Kx ) = ex .
Variance of Kx

Var (Kx ) = E Kx2 (E [Kx ])2


where

E Kx2 = ∑ k 2 k j qx
k =0
∞ ∞
= 2 ∑ k k px ∑ k px .
k =1 k =1

Ruhiyat (Department of Mathematics IPB) Actuarial Mathematics Bogor, 2020 20 / 20

You might also like