STAT 3908: Credibility Theory and Loss

Distributions
Greatest Accuracy Credibility Theory

Tutorial 2

1 / 16
Bayesian framework

I Prior distribution: fΘ (θ), Θ may stand for the risk profile of

individuals and is unobservable
I X : stands for claim size or claim number
I Conditional pdf/pmf of X given Θ: fX |Θ (x|θ)
I Hypothetical mean: µ(θ) = E[X |Θ = θ] = θ
I Joint pdf: fX ,Θ (x, θ)
R
I Marginal pdf/pmf of X : fX (x) = Θ fX ,Θ (x, θ)dθ
I Conditional pdf of Θ given X :

fX ,Θ (x, θ)
fΘ|X (θ|x) = ∝ fX ,Θ (x, θ) = fX |Θ (x|θ)fΘ (θ)
fX (x)

2 / 16
Bayesian approach to credibility

Task:
to “predict” the value of xn+1 given observations X = x.

In general, it is assumed that

I conditioned on Θ = θ, the loss amounts X1 , . . . , Xn , Xn+1 are
independent.
I Also, we assume that the prior pdf/pmf π(θ) and the
condition pdf/pmf fXj |Θ (xj |θ) are known.

Predictive distribution of Xn+1 (hard to obtain!)

Z
fXn+1 |X (xn+1 |x) = fXn+1 |Θ (xn+1 |θ)πΘ|X (θ|x)dθ,

where πΘ|X (θ|x) is the posterior distribution.

3 / 16
Two easy approaches to define the premium

The Pure Premium

1. Use fXn+1 |Θ (xn+1 |θ) to find µ(θ), i.e.,
Z
µ(θ) = E[Xn+1 |Θ = θ] = xn+1 · fXn+1 |Θ (xn+1 |θ)dxn+1

2. Replace θ by Θ, we have µ(Θ)

3. Calculate E[µ(Θ)], i.e.,
Z
E[µ(Θ)] = E[E[Xn+1 |Θ]] = µ(θ) · π(θ)dθ.

4 / 16
The Bayesian Premium
Z
E[Xn+1 |X = x] = πΘ|X (θ|x)µ(θ)dθ= E[µ(Θ)|X = x]

Remark: Computation will be complex if we use

Z
E[Xn+1 |X = x] = xn+1 · fXn+1 |X (xn+1 |x)dxn+1 .

5 / 16
Problem 1
The conditional distribution of the severity X given the risk
parameter Θ is

fX |Θ (x|θ) = θ2 xe −θx , x > 0.

Find the hypothetical mean.

6 / 16
Solution 1
Notice that
x
!2 − 1
1 − x
1 x 2−1 e θ
fX |Θ (x|θ) = θ2 xe −θx = 1
x 2−1 e θ = .
1 2


θ Γ(1) θ

Thus,  
1
X |Θ = θ ∼ Gamma 2, .
θ
The hypothetical mean is
2
µ(θ) = E[X |Θ = θ] = .
θ

7 / 16
Problem 2
You are given:
1. The Pareto distribution has the CDF as

F (x) = 1 − (θ/x)α , x ≥ θ,

where α is the shape parameter and θ is the scale parameter.

2. The prior distribution of the parameter Θ has probability
1
density function π(θ) = 2 , 1 < θ < ∞.
θ
3. Given Θ = θ, claim size follows a Pareto distribution with
parameters α = 2 and θ.
4. A claim of 3 is observed.
Calculate the posterior probability that Θ exceeds 2.

8 / 16
Solution 2
The Pareto distribution has the CDF as
 α
θ
F (x) = 1 − , x ≥ θ.
x

Note that
αθα
fX |Θ (x|θ) = , θ ≤ x.
x (α+1)
Then
2θ2
fX |Θ (3|θ) = , θ ≤ 3.
33
Given an observation of X = 3, the posterior distribution satisfies
that
2θ2 1
πΘ|X (θ|3) ∝ · , 1 < θ ≤ 3.
27 θ2

9 / 16
Then, the normalizing factor for this to be a proper density
function is
1 27
C = R3 2 R∞ = .
4
1 27 dθ + 3 0dθ
Therefore,

2θ2 1 1
πΘ|X (θ|3) = C · · 2 = , 1 < θ ≤ 3.
27 θ 2
Then, Z 3
1 1
P(Θ > 2|X = 3) = dθ + 0 = .
2 2 2

10 / 16
Problem 3
You are given:

Number Claim Count Probabilities

Class of Insureds 0 1 2 3 4
1 3000 1/3 1/3 1/3 0 0
2 2000 0 1/6 2/3 1/6 0
3 1000 0 0 1/6 2/3 1/6

A randomly selected insured has one claim in year 1. Determine

the expected number of claims in year 2 for that insured.

11 / 16
Solution 3
From the table, the prior distributions are
3000 1 2000 1 1000 1
πΘ (1) = = , πΘ (2) = = , πΘ (3) = = .
6000 2 6000 3 6000 6
Therefore, the posterior density given that X1 = 1 is

fX |Θ (1|θ)πΘ (θ)
πΘ|X1 (θ|1) = P3 1 .
f
j=1 1X |Θ (1|j)πΘ (j)

Thus,
(1/3)(1/2) 3
πΘ|X1 (1|1) = = ;
(1/3)(1/2) + (1/6)(1/3) + (0)(1/6) 4

(1/6)(1/3) 1
πΘ|X1 (2|1) = = ;
(1/3)(1/2) + (1/6)(1/3) + (0)(1/6) 4

12 / 16
 (0)(1/6)
πΘ|X1 (3|1) = = 0.
(1/3)(1/2) + (1/6)(1/3) + (0)(1/6)
Therefore, the expected number of claims in year 2, denoted by
X2 , for that insured is

E[X2 |X1 = 1] = E[X2 |Θ = 1]πΘ|X1 (1|1) + E[X2 |Θ = 2]πΘ|X1 (2|1)

+E[X2 |Θ = 3]πΘ|X1 (3|1)
        
1 1 1 3
= (0) + (1) + (2)
3 3 3 4
        
1 2 1 1
+ (1) + (2) + (3)
6 3 6 4
= 1.25

13 / 16
Problem 4
Suppose that Θ has pdf π(θ), θ > 0, and Θ1 has pdf
π1 (θ) = π(θ − α), θ > α > 0. If, given Θ1 , X is Poisson
distributed with mean Θ1 , show that X has the same distribution
as Y + Z , where Y and Z are independent, Y is Poisson
distributed with mean α, and Z |Θ is Poisson distributed with
mean Θ.

14 / 16
Solution 4
Note that
e −θ θx e −θ θx
Z Z
fX (x) = π1 (θ)dθ = π(θ − α)dθ,
x! x!

e −α αy
fY (y ) = ,
y!
e −θ θz
Z
fZ (z) = π(θ)dθ.
z!
Let W = Y + Z . Then,

15 / 16
fW (w )
Xw
= fY (y )fZ (w − y )
y =0
w
e −α αy
e −θ θw −y
X Z
= π(θ)dθ
(w − y )!
y!
y =0
Z −(α+θ) w   y  w −y
e (α + θ)w X w α θ
= π(θ)dθ
w! y α+θ α+θ
y =0
Z −(α+θ)
e (α + θ)w
= π(θ)dθ,
w!
with the last line following because the sum contains binomial
probabilities. Let r = α + θ and so
Z −r w
e r
fW (w ) = π(r − α)dr = fX (w ).
w!

16 / 16