SAS 4291: Loss Models and Credibility Theory

Course Purpose:
The purpose of this course is to equip students with the tools for modelling, estimation and interpretation
of quantitative risk models insurance loss models.

Course Objectives:
At the end of the course, the students should be able to:

1. Describe the properties of the statistical distributions which are suitable for modelling individual
and aggregate losses.

2. Construct risk models involving frequency and severity distributions and calculate the moment
generating function and the moments for the risk models both with and without simple reinsurance
arrangements.

Course Description
Effects of coverage modifications; Deductible, Policy limit, Coinsurance, Effects of inflation, Effects of
deductible on claim frequency, Coverage modifications and stop-loss reinsurance.

Reinsurance: Introduction- proportional reinsurance arrangements - excess of loss reinsurance for prior
and re-insurer- proportional reinsurance; log-normal distribution and examples -normal distribution and
examples - inflation - estimation - policy excess.

Credibility theory: Introduction - credibility: the credibility premium formula the credibility prior.
Limited fluctuation (classical) credibility, full and partial credibility. Buhlmann and Buhlmann-Straub
credibility.

Bayesian credibility: introduction - the Poisson/gamma model- numerical illustrations of the

Poisson/gamma model - the normal/normal model - dissuasion of the Bayesian approach to credibility.
Empirical Bayes credibility theory: Empirical Bayes credibility theory Models.

Teaching Methodologies
Lecture Method covering, Independent Study and Group discussions (3 hours weekly), Case Studies,
Simulations, Computer Tutorials (R-software)

Instructional Materials/Equipment
Texts, audio and video cassettes, computer software, case studies

Course Assessment
The assessment will be conducted by CATs (Continuous Assessment Tests), regular assignments and
a final Examination at end of the semester. The composition for assessment shall be as follows: 30%
Assignments and CATs and 70% end of semester examinations. Coursework: 30%
Final Written Examination: 70%
Total: 100%

1
Prerequisite
SAS 3290 Actuarial Risk Theory.

Core Text:
Tse Y., Non-life Actuarial Models: Theory, Methods and Evaluation, Cambridge University Press,
ISBN: 9780521764650, 2009.

Other Texts:
1 Buhlmann H., Gisler A., A Course in Credibility Theory and its Applications, Springer, ISBN:
9783540292739, 2006.

2 Klugman S., et al, Loss Models, Solutions Manual: From Data to Decisions, 2nd edition, Wiley,
ISBN: 9780471227625, 2004.

3 Ross S.M., Introduction to Probability Models, 10th edition, Academic Press, ISBN:
9780123756879,2006

5 Bühlmann, Hans, and Alois Gisler. 2005. A Course in Credibility Theory and Its Applications.
ACTEX Publications.

6 Klugman, Stuart A., Harry H. Panjer, and Gordon E. Willmot. 2012. Loss Models: From Data to
Decisions. John Wiley & Sons.

7 Tse, Yiu-Kuen. 2009. Nonlife Actuarial Models: Theory, Methods and Evaluation. Cambridge
University Press.

Course Journals
(1) Journal of the American Statistical Association, ISSN: 0612-1459.

(2) A survey of Credibility Theory, ISSN: 1568-4539.

(3) Scandinavian Actuarial Journal. Published/Hosted by Taylor and Francis Group. ISSN: 0346-
1238

2
 Contents

1 Effects of coverage modifications 4

1.1 Deductibles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.1 Ordinary Deductibles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 The Loss Elimination Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.1.3 Franchise Deductibles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2 The Effect of Inflation for Ordinary Deductibles . . . . . . . . . . . . . . . . . . . . . . 19
1.3 Policy Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 Coinsurance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.5 Combinations of Coinsurance, Deductibles, Limits, and Inflations . . . . . . . . . . . . 23
1.6 The Impact of Deductibles on Claim Frequency . . . . . . . . . . . . . . . . . . . . . . 25

2 Aggregate Loss Models 28

2.1 Stop Loss Insurance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 Individual Policy Modification Impact on Aggregate Losses . . . . . . . . . . . . . . . . 35

3
 1 Effects of coverage modifications

Up to now we have assumed that the insurer pays the totality of any loss as given by the random
variable X. In practice, the insurer often only covers part of the loss, leaving the insured to pay the
remainder. The prime motivation is to make the insured party partially responsible, so that they have an
interest in taking steps to avoid loss.

This has major implications when we look at the statistical problem of inferring details about loss
distributions from the data furnished by insurers on their claims experience. In practice, this data will
give the amounts actually paid on claims, rather than the actual losses. In order to estimate loss
distributions, one needs to understand clearly the relationship between the amount of the loss and the
amount actually paid under the common types of modifications.

To reduce risks and/or control problems of moral hazard,1 insurance companies often modify the
policy coverage. Examples of such modifications are deductibles, policy limits, and coinsurance.
These modifications change the amount paid by the insurance companies in case of a loss event. For
example, with deductibles the insurer does not incur any payment in a loss event if the loss does not
exceed the deductible. Thus, we need to distinguish between a loss event and a payment event. A loss
event occurs whenever there is a loss, while a payment event occurs only when the insurer is liable to
pay for (some or all of) the loss.

In this lecture, we study the distribution of the loss to the insurer when there is coverage modification.
To begin with, we define the following notations:

1. X = amount paid in a loss event when there is no coverage modification, also called the ground-
up loss

2. XL = amount paid in a loss event when there is coverage modification, also called the cost per
loss

3. XP = amount paid in a payment event when there is coverage modification, also called the cost
per payment

Thus, X and XP are positive and XL is nonnegative. Now we consider some coverage modifications and
their effects on the loss-amount variable XL and the payment-amount variable XP .

1.1 Deductibles
Under this arrangement, the insurer only pays the losses that are above some amount d fixed in advance.
The purchaser of the insurance pays for the first d units of loss, and of course if the loss is less than d,
the insurer pays nothing, and the insured is fully responsible. This has an added advantage to the insurer
of preventing an undue expenses involved in processing small claims.

1.1.1 Ordinary Deductibles

For an insurance policy, in order for a claim to be paid, a threshold d must be exceeded. That is, the
ground-up loss X must exceed d. In this case, the insurance pays the policyholder the amount X − d.
For loss amount less than d the insurance pays 0, an insured is fully responsible. This is referred to as
1 Moral hazard refers to the situation in which the insured behaves differently from the way he would behave if he were
fully exposed to the risk, such that the risk of burden is transferred to the insurer. For example, in vehicle insurance, an insured
may drive less carefully, knowing that the insurer will pay for the damages in accidents.

4
ordinary deductible. The amount paid by the insurance is called the cost-per loss2 or left censored
and shifted variable.

If the original severity distribution is given by the random variable X, and there is a deductible of d, the
amount paid by the insurer in a loss event, XL is given by
(
0, X ≤ d,
XL = max{X − d, 0} = (X − d)+ = (1.1)
X − d, X > d.

The cost-per loss XL is an example of a mixed random variable: For X ≤ d, XL is a discrete random
variable with probability at 0 given by

Pr (XL = 0) = Pr(X < d) = FX (d).

It has a probability mass at point 0 of FX (d).

The continuous part of XL has a density function given by;

fXL (x) = fX (x + d), for x > 0.

Therefore; (
FX (d), x=0
fXL (x) =
fX (x + d), x > 0
The cdf of XL is given by

FXL (x) = Pr(XL ≤ x) = Pr(X − d ≤ x) = FX (x + d), x ≥ 0

The sdf of XL is given by

SXL (x) = 1 − FXL (x) = 1 − FX (x + d), x≥0

SXL (x) = 1 − FX (x), when x = 0
SXL (x) = 1 − FX (x + d) = SX (x + d), x>0

From this, it follows that

SXL (x) = SX (x + d), for x ≥ 0.
Example 1.1
Determine the pdf, cdf, and the sdf for XL if the ground-up loss amount function has an
exponential distribution with mean θ1 and an ordinary deductible of d.
1
Solution. Exponential distribution with mean θ is given by

fX (x) = θ e−θ x , x > 0.

Recall XL is an example of a mixed random variable.

For the discrete part XL is given by

Pr(XL = 0) = Pr(X ≤ d) = FX (d) = 1 − fX (d) = 1 − e−θ d , x=0

For continuous part XL is given by

fXL (x) = fX (x + d) = θ e−θ (x+d) , x>0

2 The random variable XL is also referred to as the claim amount paid per-loss event.

5
Therefore;
(
1 − e−θ d , x=0
fXL (x) = −θ (x+d)
θe , x > 0.

The cdf of X is given by

Z x
FX (x) = θ e−θt dt = 1 − e−θ x
0
FXL (x) = Pr (XL ≤ x) = Pr (X − d ≤ x) = Pr(X ≤ x + d) = FX (x + d), x>0
Z x+d
FX (x + d) = θ e−θt dt = 1 − e−θ (x+d) , x>0
0

When x = 0; FX (d) = 1 − e−θ d and FX (x + d) = 1 − e−θ (x+d) , x > 0.

Therefore;
(
1 − e−θ d , x=0
FXL (x) = −θ (x+d)
1−e , x > 0.

The survival function of X is given by

SX (x) = 1 − FX (x) = Pr(X > x)

(
1 − FX (d), x=0
SXL (x) = 1 − FXL (x) =
1 − FX (x + d), x > 0
(
SX (d), x=0
=
SX (x + d), x > 0

Therefore;
(
e−θ d , x=0
SXL (x) = −θ (x+d)
e , x > 0.

Example 1.2
Let X ∼ exp (θ ). Calculate fxL , SxL FxL and hxL .

Solution.
1 − x+d
fX L (x) = fX (x + d) = e θ
θ
x+d
FX L (x) = FX (x + d) = 1 − e− θ

x+d
SX L (x) = 1 − FX L (x + d) = e− θ

1 θ x+d
fX L (x) e 1
hX L (x) = = θ x+d =
SX (x) e− θ θ

6
Example 1.3
Determine similar quantities for a Pareto distribution with α = 3 and θ = 2, 000 for an ordinary
deductible of 500.

Solution. The pdf and cdf of the Pareto distribution with parameters (α, γ) are tabulated as
follows:
αγ α
fX (x) = , x≥0
(x + γ)α+1
 α
γ
FX (x) = 1 −
x+γ

We know that (
FX (d), x=0
fXL (x) =
fX (x + d), x > 0

Pr(XL = 0) = Pr(X ≤ d) = FX (d)

Z d Z d
αγ α
FX (d) = fX (x) dx = dx
0 0 (x + γ)α+1
Z d
= αγ α (x + γ)−(α+1) dx
0
" #
−(α+1)+1 d
α (x + γ)
= αγ
−α 0
−α −α
α
 
= −γ (d + γ) − γ
γα
= 1−
(d + γ)α
 α
γ
= 1−
d +γ

Therefore;  α
γα γ
FX (d) = 1 − = 1−
(d + γ)α d +γ
Hence;  3  3
2000 2000
FX (d) = 1 − = 1− = 1 − 0.512 = 0.488
500 + 2000 25000
Thus 
0.488, x=0
fXL (x) = 3(2000) 3
 , x>0
(x + 2500)4
The cumulative distribution function of XL is given by
(
FX (d), x=0
FXL = FX (x + d), x ≥ 0 =
FX (x + d), x > 0

Hence 
0.488, x=0
FXL (x) = (2000)3
1 − , x>0
(x + 2500)3

7
 The survival function SXL (x) is given by

SXL (x) = 1 − FXL (x) = 1 − FX (x + d), x ≥ 0

(
1 − FX (d), x=0
=
1 − FX (x + d) = SX (x + d), x > 0

1 − 0.488, x=0
= (2, 000)3
 x>0
(2500 + x)3


undefined, x=0
fX (x) 
hXL (x) = L = 3(2000)3 (2500 + x)3 3
SXL (x)  · = , x>0
(2500 + x)4 (2000)3 2500 + x

The kth moments of XL are given by

h i
E (XL )k = ∑ (x − d)k pX (x)
x>d

If X is discrete and h i Z∞
E (XL )k = (x − d) fX (x)dx
d
If X is continuous. Moreover, for the continuous case, if k = 1, we have
Z ∞ Z ∞
E(XL ) = (x − d) fX (x)dx = [1 − FX (x)]dx.
d d

The mean of XL can be computed as follows

Z ∞
E(XL ) = x fXL (x)dx
Z0 ∞
= (x − d) fX (x)dx
d
Z ∞
=− (x − d) dSX (x)
d

dF(x) d d
Note f (x) = = (1 − SX (x)) = − SX (x)
dx dx dx
Using integration by parts:
du dv d
u = x−d ⇒ = 1, = SX (x) ⇒ v = SX (x)
dx dx dx
Hence
 ∞ Z ∞ 
E(XL ) = − (x − d)SX (x) − SX (x)dx
d d
 Z ∞ 
= − 0− SX (x) dx
d
Z ∞
= SX (x)dx.
d

It follows that E(XL ) ≤ E(X) since

Z ∞ Z ∞ Z ∞
E(XL ) = SX (x)dx = [1 − FX (x)]dx ≤ [1 − FX (x)] dx.
d d 0

8
Note that with the presence of deductibles, the number of payments is fewer than the losses since losses
with amount less than or equal to the deductible will result in no payments.

In the cost per loss situation, all losses below or at the deductible level are recorded as 0. We next
examine the situation where all losses below or at the deductible level are completely ignored and not
recorded in any way. This situation is represented by the random variable.
(
undefined, X ≤ d
XP = (XL |XL > 0) = (XL |X > d) =
X −d X > d.

The random variable XP is called the excess loss variable,3 the cost-per payment, or the left truncated
and shifted variable. It is defined only when there is a payment, i.e., when X > d. It is a conditional
random variable, defined as XP = X − d|X > d. XP follows a continuous distribution if X is continuous,
and its pdf is given by

fXP (x) = Pr (XL |XL > 0) = Pr (XL |X > d)

Pr (XL ∩ X > d)
=
Pr(X > d)
fX (x + d)
= , x>0
SX (d)
FXP (x) = Pr (XL |XL > 0 ≤ x) = Pr (XL |X > d ≤ x)
FX (x + d) − FX (d)
= , x>0
SX (d)
SX (x + d)
SXP (x) = , x>0
SX (d)
fX (x + d)
hXP (x) = = hX (x + d), x>0
SX (x + d)
Note that in empirical applications, only data on payments are available. When the loss X is less than
or equal to d, a claim is not made and the loss information is not captured. Thus, for a policy with
deductible, only XP is observed, and some information about X and XL is lost.

Example 1.4
Determine the pdf, cdf, sdf and the hazard rate function for XP if the ground-up loss amount
function has an exponential distribution with mean θ1 and an ordinary deductible of d.
Solution. From the Tables
fX (x) = θ e−θ x , x>0
Thus;
fX (x + d)
fXP (x) = , x>0
SX (d)
fX (x + d) = θ e−θ (x+d)
FX (x + d) = 1 − e−θ (x+d)
 
SX (d) = 1 − FX (d) = 1 − 1 − e−θ d = e−θ d

Thus;
θ e−θ (x+d)
fXP (x) = = θ e−θ x , x>0
e−θ d

3 We refer to XP the amount paid per-payment event.

9
The CDF of FXP (x) is given by

FX (x + d) − FX (d) 1 − e−θ (x+d) − (1 − e−θ d )

FXP (x) = =
SX (d) e−θ d
e−θ d 1 − e−θ x
 
= = 1 − e−θ x
e−θ d
The survival function SXP (x) is given by

SX (x + d)
SXP (x) = , x>0
SX (d)
SX (x + d) = e−θ (x+d) , SX (d) = e−θ d
SX (x + d) e−θ (x+d) e−θ x · e−θ d
∴ SXP (x) = = = = e−θ x , x>0
SX (d) e−θ d e−θ d

Alternatively;
SXP (x) = 1 − FXP (x) = 1 − (1 − e−θ x ) = e−θ x .
The hazard function hXP (x) is given by

fX (x + d)
hXP (x) = = hX (x + d), x>0
SX (x + d)
θ e−θ (x+d)
= =θ
e−θ (x+d)

Example 1.5
Let X ∼ exp (θ ). Derive fX P , SX P FX P and hX P .

Solution.
1 1 d
1 − θ (x+d)
fX (x + d) e 1
θ e θ (x+d) eθ 1 −x
fX P = = θ = −d = 1 = e θ
SX (d) 1 − FX (d) e θeθ θ (x+d) θ
Observe that fX L is exactly the same as fX (x). This reflects the memory-less property of the
exponential distribution. Given that the loss is already of size d, the probability of the total loss
being x + d is the same as the probability of a loss of size x. Basically, conditioning has no effect.
when we compute the other quantities.
1
SX (x + d) e− θ (x+d) d 1 x
SX P = = d = e− θ − θ (x+d) = e− θ = SX (x)
SX (d) −
e θ
Note that we could have arrived at the same conclusion by simply seeing that fX P is exactly the
same as fX (x). The calculation confirms this, but is unnecessary.
 −d

− θ1 (x+d)
FX (x + d) − FX (d) 1 − e − 1 − e θ
x
FX P = = d = 1 − e− d = FX (x)
SX (d) e− θ
fX (x+d) 1 −x+d
f P (x) SX (d) θe θ 1
hX P (x) = X = = =
SX P (x) SX (x+d) − x+d θ
SX (d) e θ

10
Example 1.6
Derive the following quantities fXP (x), SXP (x), FXP (x) and hXP (x) for a Pareto distribution with
α = 3 and θ = 2, 000 for an ordinary deductible of 500.

Solution. From the Tables;

 α
αγ α γ
fX (x) = , x ≥ 0, FX (x) = 1 −
(x + γ)α+1 x+γ

Therefore;
fX (x + d)
∴ fXP (x) = , x>0
SX (d)
αγ α
fX (x + d) = , x>0
(x + d + γ)α+1
γα
FX (x + d) = 1 − x≥0
(x + d + γ)α
γα
FX (d) = 1 − x=0
(d + γ)α
∴ SX (d) = 1 − FX (d)
 
γα γα
= 1− 1− =
(d + γ)α (d + γ)α

Thus,
αγ α (d + γ)α α(d + γ)α
fXP (x) = · =
(x + d + γ)α+1 γ α (x + d + γ)α+1
Hence;

3(2, 000)3 (2, 000 + x + 500)−4 3(2500)3

fXP (x) = = .
(2, 000)3 (2, 000 + 500)−3 (2500 + x)4

SX (x + d)
SXP (x) = , x>0
SX (d)
γα
SX (x + d) = 1 − FX (x + d) =
(x + d + γ)α
γ α (d + γ)α (d + γ)α
SXP (x) = · = , x>0
(x + d + γ)α γα (x + d + γ)α
 3
2500
∴ SXP (x) =
2500 + x
 3
FX (x + d) − FX (d) 2500
FXP (x) = 1 − SXP (x) = = 1−
SX (d) 2500 + x
fX (x + d) αγ α (x + d + γ)α α 3
hXP (x) = hX (x + d) = = · = =
SX (x + d) (x + d + γ) α+1 γ α x+d +γ 2500 + x

11
The mean of XP , is called the mean excess loss, and is given by the following formula
Z ∞
E(XP ) = x fXP (x)dx
0
Z ∞  
fX (x + d)
= x dx
0 SX (x)
R∞
x fX (x + d)dx
= 0
SX (d)
R∞
(x − d) fX (x)dx
= d
SX (d)
E(XL )
= .
SX (d)
Note that we can also express XP as XP = XL |XL > 0. Now using conditional expectation, we have

E(XL ) = E(XL |XL > 0)Pr(XL > 0) + E(XL |XL = 0)Pr(XL = 0)

= E(XL |XL > 0)Pr(XL > 0)
= E(XP )Pr(XL > 0)

which implies
E(XL ) E(XL ) E(XL ) E(XL )
E(XP ) = = = = > E(XL )
Pr(XL > 0) 1 − FX (d) SXL (0) SX (d)
That is, the mean excess loss is larger than the expected amount paid per loss event.

Theorem 1.1
For an ordinary deductible d

(1) The expected cost per Loss is: E (XL ) = E (X) − E (X ∧ d)

E (X) − E (X ∧ d) E (XL )
(2) The expected cost per payment is: E (XP ) = =
SX (d) SX (d)

Another random variable of interest; which is associated with an ordinary deductible, d, above is
(
X, if x < d,
X ∧ d = min{X, d} =
d, if x ≥ d.

where X ∧ d is the limited loss variable.

By looking separately at the case where X is less than or greater than d, it follows that, in general,

X = (X − d)+ + (X ∧ d)

So that

E(X − d)+ = E(X) − E(X ∧ d) (1.2)

and for continuous X, it follows from Eqn. (1.2) or by calculating directly that
Z d Z d
E(X ∧ d) = x fX (x)dx + dSX (d) = SX (x)dx.
0 0

Also
E(XL ) E(X) − E(X ∧ d)
E(XP ) = =
1 − FX (d) 1 − FX (d)

12
Example 1.7
For an insurance:

(i) Losses have density function

(
0.02x, 0 < x < 10
f (x) =
0, otherwise

(ii) The insurance has an ordinary deductible of 4 per loss.

(iii) Y P is the claim payment per payment random variable,

Calculate E Y P .
 

Solution.
Z 4 Z 4
SX (4) = 1 − fX (x) = 1 − 0.02x dx
0 0
4
= 1 − 0.01x2 = 0.84
0

fX (y + 4) 0.02(y + 4)
fY P (y) = = = 0.0238(y + 4)2
SX (4) 0.84

Z 6
P
E(Y ) = y(0.0238(y + 4)) dy
0
4y2 6
 3 
y
= 0.0238 + = 3.4272
3 2 0

Example 1.8
Determine the four expectations for the Pareto distribution with α = 3 and θ = 2, 000 for an
ordinary deductible of 500.
Solution.
Z d Z 500  α
θ
E(X ∧ d) = SX (x) dx = dx
0 0 x+θ
Z 500  3 Z 500
2, 000
= dx = 2, 000 (x + 2, 000)−3 dx
3
0 x + 2, 000 0

2, 0003   500 2, 0003

 
−2 1 1
= (x + 2, 000) = −
−2 0 −2 (2500)2 2, 0002
= 360
 3
2, 000
FX (500) = 1 − = 0.488,
2, 000 + 500
"  2 #
2, 000 2, 000
E(X ∧ 500) = 1− = 360
2 2, 000 + 500
E(XL ) = E(X) = E(X ∧ d)

13
 with E(X) = 1000 we have, for the ordinary deductible, the expected cost per loss is, 1, 000 −
360 = 640. While the expected cost per payment is

E(XL ) 640 640 640

E(XP ) = = = = = 1250
SX (d) 1 − FX (500) 1 − 0.488 0.512

Example 1.9
X takes the values 100, 200, 300, 400 with probabilities 0.4, 0.3, 0.2, 0.1 respectively. Describe
the distributions of (X − d)+ and X ∧ d, for d = 230. Find E(X − 230)+ and E(X ∧ 230).

Solution. (X − 230)+ takes the values 0 with probability 0.7, (first two cases, 100 and 200),
70 with probability 0.2, and 170 with probability 0.1, while X ∧ 230 takes the value 100 with
probability 0.4, 200 with probability 0.3 and 230 with probability 0.3. Calculating directly

E(X ∧ 230) = (100 × 0.4) + (200 × 0.3) + 230 × 0.3 = 169,

E(X − 230)+ = 70 × 0.2 + 170 × 0.1 = 31.

It is often convenient to compute E(X − d)+ from the formula:

E(XL ) = E(X − d)+ = E(X) − E(X ∧ d).

This is particularly true when X is infinite and discrete.

We can compute E(X ∧ d) by a finite sum, as opposed to the infinite series involved in computing
E(X − d)+ directly.

Example 1.10

Suppose X ∼ Geom(p). Find E(X − 21 )+ .

Solution. Since X is always greater than 1/2 unless it is equal to 0, we know that X ∧ 1/2 takes
the values 0 with probability 1 − p and the values 1/2 with probability p. So

E(X ∧ 1/2) = p/2.

p
Since E(X) = 1−p , we conclude

1 p p
E(X − )+ = − .
2 1− p 2

Question 1.1. Let X ∼ exp(λ ), and d be the amount of the deductible.

(a) What is the df of XL = (X − d)+ ?

(b) What is the df of XP = X − d|X > d?

(c) Find the pdf of XP and E(XP ).

Question 1.2. The pdf of the claim severity X is fX (x) = 0.02x, for 0 ≤ x ≤ 10. An insurance policy
has a deductible of d = 4, calculate the expected loss payment in a payment event.

14
1.1.2 Loss Elimination Ratio
Loss amounts X subject to an ordinary deductible d satisfy the relation
( (
0, X ≤ d, X, X ≤ d,
X = (X − d)+ + X ∧ d = +
X − d, X > d. d, X > d.

In words, this relation says that for a policy with a deductible d, losses below d are not covered and
therefore can be covered by another policy with a limit of d. Rewriting the above relation as

X ∧ d = X − (X − d)+

we note that X ∧ d is a decrease of the overall losses and hence can be considered as savings to the
policyholder in the presence of deductibles.

The Loss Elimination Ratio (LER) is meaningful in evaluating the impact of a deductible.

Definition 1.1
The Loss Elimination Ratio (LER) is the ratio of the decrease in the expected payment with an
ordinary deductible to the expected payment without the deductible.

While many types of coverage modifications can decrease the expected payment, the term loss
elimination ratio is reserved for the effect of changing the deductible. Without the deductible, the
expected payment is E(X). With the deductible, the expected payment is E(X) − E(X ∧ d). Therefore
the loss elimination ratio is
E(X) − [E(X) − E(X ∧ d)] E(X ∧ d)
LER = = .
E(X) E(X)
provided E(X) exists.

The expected savings (due to the deductible) expressed as a percentage of the loss (no deductible at all)
is called the Loss Elimination Ratio.

For a continuous loss amounts X we have

Rd
[1 − FX (x)]dx
LER = R 0∞ .
0 [1 − FX (x)]dx

Example 1.11
Let losses be distributed by a Pareto(a = 4, q = 5000) distribution. Let d = 1000, where d is an
ordinary deductible. Compute the LER.
Solution. Recall that the formula for LER is:
E(X) − [E(X) − E(X ∧ d)] E(X ∧ d)
LER = = .
E(X) E(X)
using the information we know about the Pareto distribution, Then LER is simply
h
α−1 i
θ θ
1−α 1 − d+θ
LER = θ
α−1
 α−1
θ
= 1−
d +θ

15
  4−1
5000
= 1−
1000 + 5000
= 0.42

Example 1.12
Loss amounts X follows an exponential distribution with mean θ = 1000. Suppose that insurance
policies are subject to an ordinary deductible of 500.

Calculate the loss elimination ratio.

Solution.
E(X) − [E(X) − E(X ∧ d)] E(X ∧ d)
LER = = .
E(X) E(X)
 d

θ 1 − e− θ
=
θ
d
= 1 − e− θ
500
= 1 − e− 1000
= 0.393469

Question 1.3. Let X ∼ Pareto(2, 5) and d = 1 be the amount of the deductible.

(a) Calculate E(XP )

(b) Calculate E(X ∧ d)

(c) Calculate LER(d).

1.1.3 Franchise Deductibles

The second type of deductible that we consider is the franchise deductibles. Under this deductible
policy, the loss is fully paid when the loss exceeds the deductible.

The cost per-loss random variable is defined by

(
0, X ≤ d,
XL =
X, X > d.

and the cost per-payment is

(
undefined, X ≤ d
XP = (X|X > d) =
X, X > d.

16
The key distributional functions are
(
FX (d), x = 0
fXL (x) =
fX (x), x > d

(
FX (d), 0 ≤ x ≤ d
FXL (x) =
FX (x), x > d

(
SX (d), 0 ≤ x ≤ d
SXL (x) =
SX (x), x > d

(
0, 0<x<d
hXL (x) =
hX (x), x > d.

for the cost per-loss and

fX (x)
fXP (x) = , x>d
SX (d)


0, 0≤x≤d
FXP (x) = FX (x) − FX (d)
 , x>d
SX (d)


1, 0≤x≤d
SXP (x) = SX (x)
 , x>d
SX (d)


0,
 0<x<d
hXP (x) = hX (x), x > d



for the cost per-payment.

17
Theorem 1.2
For a franchise deductible d, we have

E(XL ) = E(X − d)+ = E(X) − E(X ∧ d) + d [1 − FX (d)]

and
E(X) − E(X ∧ d)
E(XP ) = +d
1 − FX (d)
where X ∧ d is the limited loss variable defined by
(
X, X < d
X ∧ d = min{X, d} =
d, X ≥ d

and Z d
E(X ∧ d) = SX (x)dx.
0

Proof. We will proof the results for the continuous case. We have
Z ∞ Z ∞ Z ∞
E(XL ) = x fX (x)dx = (x − d) fX (x)dx + d fX (x)dx.
d d d
= E [(X − d)+ + d [1 − FX (d)]]
= E(X) − E(X ∧ d) + d [1 − FX (d)]

E(XL ) E(X) − E(X ∧ d)

E(XP ) = = + d.
[1 − FX (d)] 1 − FX (d)

Example 1.13
Let X ∼ exponential (θ = 1000) and let d = 250. when d is an ordinary deductible, what is the
expected cost per loss and the expected cost per payment event.
 −d

Solution. we have that the E (X ∧ d) = θ 1 − e θ , so
 −d
 −d −250
E X L = E (X) − E (X ∧ d) = θ − θ 1 − e θ = θ e θ = 1000e 1000 = 778.801

−d
P E (X) − E (X ∧ d) θ e θ


E X = = −d = θ = 1000
1 − FX (d) eθ
Now what if this were a franchise deductible? Then we have

E X P = E (X) − E (X ∧ d) + d (1 − FX (d))

 −d
 −d
= θ − θ 1 − e θ + de θ

d
= (θ + d) e− θ

250
= (1250) e− 1000

= 973.501

18

E (X) − E (X ∧ d)
E XP = +d
1 − FX (d)
 −d


θ −θ 1−e θ

E XP = −d + d = θ + d = 1250
eθ
Note that for any loss X, a franchise deductible always pays at least as much as an ordinary
deductible

Example 1.14

Example 1.15
Let losses be distributed by a Pareto(a = 4, q = 5000) distribution. Let d = 1000, where d is a
franchise deductible. Compute the LER.
Solution.
E(X) − [E(X) − E(X ∧ d) + d (1 − FX (d))] E(X ∧ d) − d (1 − FX (d))
LER = =
E(X) E(X)

using the knowledge about the Pareto distribution, the LER is simply
α−1 θ

α−1
d

θ d+θ
LER = 1 − −
d +θ θ
α−1
4−1 5000

4−1
1000

5000 1000+5000
= 1− − 5000
1000 + 5000 4−1

= 0.132

Example 1.16
You are given:

(i) Losses follow an exponential distribution with the same mean in all years.

(ii) The loss elimination ratio this year is 70%.

(iii) The ordinary deductible for the coming year is 4/3 of the current deductible.

Compute the loss elimination ratio for the coming year.

Question 1.4. Determine the loss elimination ratio for the Pareto distribution with α = 3 and θ = 2, 000
with an ordinary deductible of 500.
From Example ??, we have a loss elimination ratio of 360/1,000 = 0.36. Thus 36% of losses can be
eliminated by introducing an ordinary deductible of 500.

1.2 Effect of Inflation for Ordinary Deductibles

While loss distributions are specified based on current experience and data, inflation may cause increases
in the costs. However, policy specifications (such as deductible) remain unchanged for the policy period.

19
To model the effects of inflation, we consider a one-period insurance policy and assume the rate of price
increase in the period to be r. We use a tilde to denote inflation-adjusted losses. Thus, the inflation-
adjusted loss distribution is denoted by X̃, which is equal to (1 + r)X.
For an insurance policy with deductible d, the loss in a loss event and the loss in a payment event with
inflation adjustment are denoted by X̃L and X̃P , respectively. As the deductible is not inflation adjusted,
we have Inflation increases costs, but it turns out that when there is a deductible the effect of inflation
is magnified. First, some events that formerly produced losses below the deductible will now lead to
payments. Second, the relative effect of inflation is magnified because the deductible is subtracted after
inflation. For example, suppose an event formerly produced a loss of 600. With a 500 deductible the
payment is 100. Inflation at 10% will increase the loss to 660 and the payment to 160, a 60% increase
in the cost to the insurer.

Theorem 1.3
Let loss amounts be X and let Y be the loss amounts after uniform inflation of r. That is Y =
(1 + r)X. For an ordinary deductible of d after uniform inflation of 1 + r, the expected cost per
loss is   
d
E(XL ) = (1 + r) E(X) − E X ∧ .
1+r
The expected cost per-payment is
  −1   
E(Y ) − E(Y ∧ d) d d
E(XP ) = = 1 − FX (1 + r) E(X) − E X ∧ .
1 − FY (d) 1+r 1+r

Example 1.17
Determine the effect of inflation at 10% on an ordinary deductible of 500 applied to an
exponential distribution with mean 1000.

Question 1.5. Determine the effect of inflation at 10% on an ordinary deductible of 500 applied to a
Pareto distribution with α = 3 and θ = 2, 000.

1.3 Policy Limit

For an insurance policy with a policy limit, the insurer compensates the insured up to a pre-set amount,
say, u. If a policy has a policy limit but no deductible, then the amount paid in a loss event is the same
as the amount paid in a payment event.

If a policy has a limit u, then the insurer will pay the full loss as long as the losses are less than or equal
to u, otherwise, the insurer pays only the amount u. Thus, the insurer is subject to pay a maximum
covered loss of u. Let XU denote the claim amount paid for a policy with a policy limit u. Then,
(
X, for x ≤ u,
XU = min{X, u} = (X ∧ u) =
u, for X > u.

We call Xu the limited loss random variable.

Just as in the case of a deductible, XU has a mixed distribution, with the discrete part.

pXU (u) = Pr(X > u) = 1 − FX (u).

and a continuous part given by

fXU (x) = fX (x), x < u.

20
Moreover, fX (x) = 0 for x > u. The cdf of XU is
(
FX (x), x < u,
FXU (x) = Pr(XU ≤ x) =
1, x ≥ u.

Example 1.18
Ru
Show that: E(XU ) = 0 SX (x)dx.
Solution. First note that
XU = u + Z
where (
X − u, X ≤ u
Z=
0, X > u.
Thus,
Z u
E(XU ) = u + (x − u) fX (x)dx
0
Z u
= u + [(x − u)FX (x)]u0 − FX (x)dx
0
Z u
= [1 − FX (x)]dx
0

Theorem 1.4
Let E(XU ) be the expected cost before inflation. Suppose that the same policy limit applies after
an inflation at rate r. Then the after inflation expected cost is given by
 
u
E((1 + r)X ∧ u) = (1 + r)E X ∧ .
1+r

Proof. See the proof of Theorem 1.2

Example 1.19
Losses follow a Pareto distribution with parameters α = 2 and θ = 1000. For a coverage with
policy limit of 2000, find the pdf and the cdf of the limited loss random variable.
Solution. Recall that
αθ α 2(1000)2
fX (x) = =
(x + θ )α+1 (1000 + x)3
and  α  2
θ 1000
FX (x) = 1 − = 1− .
θ +x 1000 + x
Thus, 
1
9, x = 2000


2(1000)2
fXU (x) = , x < 2000
 (1000+x)3

0, x > 2000
and
1000 2
(

1− 1000+x , x < 2000
FXU (x) =
1, x ≥ 2000.

21
 Example 1.20
Losses follow a Pareto distribution with parameters α = 2 and θ = 1000. Calculate the expected
cost for a coverage with policy limit of 2000.
Solution. Recall that  α
θ
S(x) = .
θ +x
Thus, 2
Z 2000 Z 2000 
1000
E(X ∧ 2000) = S(x)dx = dx = 666.67
0 0 1000 + x

Example 1.21
Losses follow a Pareto distribution with parameters α = 2 and θ = 1000. For a coverage with
policy limit 2000 and after an inflation rate of 30%, calculate the after inflation expected cost.
Solution. We have
  " !#
2000 2000 1000
E(1.3X ∧ 2000) = 1.3 X ∧ = 1.3 1− = 1575.76.
1.3 2−1 1000 + 2000
1.3

Example 1.22
A jewelry store has obtained two separate insurance policies that together provide full coverage.
You are given:

(i) The average ground-up loss is 11,100.

(ii) Policy A has an ordinary deductible of 5,000 with no policy limit.

(iii) Under policy A, the expected amount paid per loss is 6,500.

(iv) Under policy A, the expected amount paid per payment is 10,000.

(v) Policy B has no deductible and a policy limit of 5,000.

Given that a loss occurred, determine the probability that the payment under policy B is 5,000.
Solution. Let X denote the ground-up loss random variable. By (i), E(X) = 11, 100. By (ii) and
(iii), we have
E(XL ) = E(X) − E(X ∧ 5000) = 6, 500.
By (iv), we have
E(X) − E(X ∧ 5000)
E(XP ) = = 10, 000.
1 − FX (5000)
Thus,
6500
= 10, 000 =⇒ FX (5, 000) = 0.35.
1 − FX (5000)
For policy B, a payment of 5000 will occur if X ≥ 5000. Hence,

Pr(X ≥ 5000) = 1 − FX (5000) = 1 − 0.35 = 0.65

Question 1.6. Losses follow an exponential distribution with mean θ1 . For a coverage with policy limit
u, find fXU (x), FXU (x), and E(X ∧ u).

22
Question 1.7. Losses follow an exponential distribution with mean θ1 . For a coverage with policy limit
u and with an inflation at rate r, find the expected cost.
Question 1.8. Losses are distributed uniformly between 0 and 100. A insurance policy which covers
the losses has an upper limit of 80. Find the expected cost.
Question 1.9. An insurance company offers two types of policies: Type Q and Type R. Type Q has no
deductible, but has a policy limit of 3000. Type R has no limit, but has an ordinary deductible of d.
Losses follows a Pareto distribution with α = 3 and θ = 2000.

Calculate the deductible d such that both policies have the same expected cost per loss.

1.4 Coinsurance
An insurance policy may specify that the insurer and insured share the loss in a loss event, which is
called coinsurance. We consider a simple coinsurance policy in which the insurer pays the insured a
fixed portion c of the loss in a loss event, where 0 < c < 1. The insurer portion of the loss is cX and the
insured portion is (1 − c)X. Under pure coinsurance the insurer pays damages whenever there is a loss.
We denote XC as the payment made by the insurer with coinsurance. The claim amount random variable
is XC = cX and its distribution function is
 x x
FXC (x) = Pr(XC ≤ x) = Pr X ≤ = FX .
c c
Its density function is given by
1 x
fXC (x) = fX .
c c
and it is also true that
E(XC ) = cE(X).

1.5 Combinations of Coinsurance, Deductibles, Limits, and Inflations

Now, suppose that a policy is subject to a coinsurance factor c, an ordinary deductible d, and a policy
limit u∗ . By convention, the coinsurance factor is applied after the application of any deductible or
limit. We define the maximum covered loss to be u = u∗ + d. Thus, in the absence of a deductible the
maximum covered loss is just the policy limit. The claim amount per loss random variable is given by

0,
 X ≤d
XL = c(X − d), d < X ≤ u

c(u − d), X > u.


In compact form, we can write

XL = c[X ∧ u − X ∧ d].
Note that c(u − d) is the maximum payment per loss.

Example 1.23
A health insurance policy has a deductible of 200, a policy limit of 5000 and a coinsurance factor
of 80%. Calculate the expected claim amount per loss event and the expected claim amount per
payment event for this policy if losses follow an exponential distribution with mean 1000.

23
 Solution. Note that u = u∗ + d = 5000 + 200 = 5200. The expected claim amount per loss is

E(XL ) = c[X ∧ u − X ∧ d]
  
1 − θu
 1
− θd
=c 1−e − 1−e
θ θ
= 0.8 1000 1 − e−5.2 − 1000 1 − e−0.2




= 650.57

The expected claim amount per payment is

E(XL ) 650.57
E(XP ) = = −0.2 = 794.61
1 − FX (200) e

For a policy subject to a coinsurance factor, c, an ordinary deductible d, policy limit u∗ , and uniform
inflation at rate r applied only to losses, the claim amount per loss is given by

d
0,
 X ≤ 1+r
XL = c[(1 + r)X − d], 1+r d u
< X ≤ 1+r
 u
c(u − d), X > 1+r


where u = u∗ + d. In compact form, we can write

XL = c[(1 + r)X ∧ u − (1 + r)X ∧ d].

Theorem 1.5
The expected value of the per-loss random variable is obtained as
    
u d
E(XL ) = c(1 + r) E X ∧ −E X ∧ .
1+r 1+r

The expected value of the per-payment random variable is obtained as

E(XL )
E(XP ) = d

.
1 − FX 1+r

Proof. We have

E(XL ) = c {E[(1 + r)X ∧ u] − E[(1 + r)X ∧ d]}

    
u d
= c (1 + r)E X ∧ − (1 + r)E X ∧
1+r 1+r
    
u d
= c(1 + r) E X ∧ −E X ∧
1+r 1+r
d


For the expected cost per payment divide the expression by 1 − FX 1+r .

To find the variance of XL , the second moment is required which is provided by the following theorem.

24
Theorem 1.6
The second moment for the per-loss random variable is

E[(XL )2 ] = [c(1 + r)]2 E[(X ∧ u∗ )2 ] − E[(X ∧ d ∗ )2 ] − 2d ∗ [E(X ∧ u∗ ) − E(X ∧ d ∗ )] .



where u∗ = u/(1 + r) and d ∗ = d/(1 + r). For the second moment of the per-payment variable,
divide this expression by 1 − FX (d ∗ ).

Example 1.24
Determine the mean and the standard deviation per loss for an exponential distribution with mean
1000 and with a deductible of 500 and a policy limit of 2500.
x

Solution. Recall that E(X ∧ x) = θ 1 − e− θ . Thus,
3000
 500

E(XL ) = E(X ∧ 3000) − E(X ∧ 500) = 1000(1 − e− 1000 ) − 1000 1 − e− 1000 = 556.74.

For the second moment of XL we have

Z 3000
x
E[(X ∧ 3000)2 ] = x2 (0.001)e− 1000 dx + 30002 e−3 = 1, 601, 703.453
0
Z 500
x
2
E[(X ∧ 500) ] = x2 (0.001)e− 1000 dx + 5002 e−0.5 = 180, 408.0209
0
E[(XL )2 ] = 1, 601, 703.453 − 180, 408.0209 − 2(500)(556.74)
= 864, 555.4321.

The variance of XL is

Var(XL ) = 864555.4321 − 556.742 = 554, 596.0045

Question 1.10. The amount of a loss has a Pareto distribution with α = 2 and θ = 5000. An insurance
policy on this loss has an ordinary deductible of 1,000, a policy limit of 10,000, and a coinsurance of
80%.

With a uniform inflation of 2%, calculate the expected claim amount per payment on this policy.
Question 1.11. Claim amounts follow a Pareto distribution with parameter αX = 3, and θX = 2000. A
policy is subject to a coinsurance rate of c. The standard deviation of the claims for the policy is
1742.24.

Determine the value of c.

Question 1.12. The loss size distribution is exponential with mean 50. An insurance policy pays the
following for each loss. There is no insurance payment for the first 20. The policy has a coinsurance
in which the insurance company pays 75% of the loss. The maximum covered loss is 100. Calculate
E(XP ).

1.6 The Impact of Deductibles on Claim Frequency

Deductibles always affect the number of payments. For example, when an imposed deductible is
increased the number of payments per period is decreased whereas decreasing deductibles results in an
increasing number of payments per period.

25
We can quantify this process, provided it can be assumed that the imposition of coverage modification
does not affect the process that produces losses or the type of individual who will purchase insurance.
For example, those who buy a 250 deductible on an auto-mobile property damage coverage may
(correctly) view themselves as less likely to be involved in an accident than those who buy full
coverage. Similarly, an employer may find that the rate of permanent disability declines when reduced
benefits are provided to employees in the first few years of employment.

Next, we examine the more general case where NL is a zero-modified distribution. recall that a
zero-modified distribution can be defined in terms of an unmodified one (as was shown in Actuarial
Risk theory). That is
0 1 − pM
pM
k = cpk , for k = 1, 2, 3, . . . , with c =
0
,
1 − p00
where p0k is the pmf of the unmodified distribution. In the case that pM
0 = 0, we call this a zero-truncated
distribution, of ZT. For other arbitrary values of pM
0 , this is a zero-modified, or ZM, distribution. The
pgf for the modified distribution is shown as

PM (z) = 1 − c + cP0 (z),

expressed in terms of the pgf of the unmodified distribution, P0 (z). When NL follows a zero-modified
distribution, the distribution of NP is established using the same relation from earlier,

PNP = PNL (1 + v(z − 1)).

Special cases:
• NL is a ZM-Poisson random variable with parameters λ and pM
0 . The pgf of NL is

1 − pM
0 1 − pM
0

λ (z−1)

PNL (z) = 1 − + e .
1 − e−λ 1 − e−λ
Thus the pgf of NP is
1 − pM
0 1 − pM
0

λ v(z−1)

PNP (z) = 1 − + e .
1 − e−λ 1 − e−λ
So the number of payments is also a ZM-Poisson distribution with parameters λ v and pM
0 . The
probability at zero can be evaluated using Pr (NP = 0) = PNP (0).

• NL is a ZM-negative binomial random variable with parameters β , r and pM

0 . The pgf of NL is

1 − pM0 1 − pM0
PNL (z) = 1 − + [1 − β (z − 1)]
1 − (1 + β )−r 1 − (1 + β )−r
Thus the pgf of NP is

1 − pM0 1 − pM0
PNL (z) = 1 − + [1 − β v(z − 1)]
1 − (1 + β )−r 1 − (1 + β )−r
So the number of payments is also a ZM-negative binomial distribution with parameters β , r and
pM
0 . Similarly, the probability at zero can be evaluated using Pr (NP = 0) = PNP (0).

Example 1.25
Aggregate losses are modelled as follows:
(i) The number of losses follows a zero-modified Poisson distribution with λ = 3 and pM
0 =
0.5.

26
 (ii) The amount of each loss has a Burr distribution with α = 3, θ = 50, γ = 1.

(iii) There is a deductible of d = 30 on each loss.

(iv) The number of losses and the amounts of the losses are mutually independent.

Calculate E(NP ) and Var(NP ).

Solution. Since NL follows a ZM-Poisson distribution with parameters λ and pM 0 , we knnow that
NP also follows a ZM-Poisson distribution, but with parameters λ v and pM
0 , where
 3
1
v = Pr(X > 30) = = 0.2441.
1 + (30/50)

Thus, NP follows a ZM-Poisson distribution with parameters λ ∗ = λ v = 0.7324 and pM

0 = 0.5.
Finally,
λ∗
 
0.7324
E(NP ) = (1 − pM
0 ) ∗ = 0.5
1 − e−λ 1 − e0.7324
= 0.7053
∗
!
λ ∗ 1 − (λ ∗ + 1)e−λ

λ∗
 
M M M
Var(NP ) = (1 − p0 ) ∗ + p0 (1 − p0 ) ∗
(1 − e−λ )2 1 − e−λ

! 2
0.7324 1 − 1.7324e−0.7324

2 0.7324
= 0.5 + 0.5
(1 − e0.7324 )2 1 − e−0.7324
= 0.72244.

27
 2 Aggregate Loss Models

We continue our discussion of modelling losses. Keep in mind that losses depend on the loss frequency
(i.e., the number of losses) and on the loss severity (i.e., the size or the amount of the loss). In this
lecture, we will concentrate our attention on aggregate loss models. An aggregate loss refers to the total
amount of losses in one period of time, which is often encountered in the analysis of a portfolio of risks
such as a group insurance policy.

Example 2.1
Let N ∼ Poisson (2) and let {X1 , X1
, ...} be an independent
identically distributed sequence with
common distribution N 15, σ 2 = 5 . Calculate E ∑Ni=1 Xi and Var ∑Ni=1 Xi

Solution.

Convention: For N = 0 we set S = 0

Example 2.2
Let S have a Poisson frequency distribution with parameter λ = 5. The individual claim amount
has the following distribution:
x fX (x)
100 0.80
500 0.16
1000 0.04
Calculate the probability that aggregate claims will be exactly 600.

Question 2.1. There are two groups of good and bad drivers in the city. The good drivers make up 70%
of the population of the drivers. The annual claim amount of each good driver is uniformly distributed
on U (0, 3000). The annual claim for each bad driver is exponentially distributed with mean θ = 5000.
As many as 100 drivers are chosen at random. Find the mean and variance of the aggregate claim.

2.1 Stop Loss Insurance

When a deductible d is applied to the aggregate loss S over a definite period, then the insurance payment
will be (
0, S≤d
(S − d)+ = max{S − d, 0} = S − (S ∧ d) =
S − d, S > d.
We refer to this payment as the stop-loss insurance. Its expected cost is called the net stop-loss
premium and can computed as
Z ∞ Z ∞ Z ∞
E[(S − d)+ ] = S(x)dx = [1 − FS (x)] dx = (x − d) fS (x)dx
d d d

in the continuous case and

E[(S − d)+ ] = ∑ (x − d) fS (x)
x>d
in the discrete case.

Note that this is identical to the discussion of ordinary deductible.

28
Example 2.3
For a certain company, losses follow a Poisson frequency distribution with mean 2 per year, and
the amount of a loss is 1, 2, or 3, each with probability 1/3. Loss amounts are independent of the
number of losses, and of each other.

An insurance policy covers all losses in a year, subject to an annual aggregate deductible
of 2. Calculate the expected claim payments for this insurance policy.
Solution. Let S denote aggregate loss before deductible.
E(S) = 2(2) = 4, since mean severity is 2.
e−2 20
fS (0) = = 0.1353, since must have 0 losses to get aggregate loss = 0.
0!
e−2 2 1
fS (1) = = 0.0902, since must have 1 loss whose size is 1 to get aggregate loss = 1.
1! 3
E(S ∧ 2) = 0 fS (0) + 1 fS (1) + 2 [1 − fS (0) − fS (1)]
= 0(0.1353) + 1(0.0902) + 2(1 − 0.1353 − 0.0902)
= 1.6392
Therefore;
E [(S − 2)+ ] = E(S) − E(S ∧ 2)
= 4 − 1.6392
= 2.3608

Example 2.4
The distribution of aggregate losses covered under a policy of stop-loss insurance is given by
1
FS (x) = 1 − 2 , x > 1. Calculate E[(S − 3)+ ].
x
Solution. We have
dx 1 ∞ 1
Z ∞
E[(S − 3)+ ] = 2
= − = .
3 x x 3 3
The following provides a simple calculation of the net stop-loss premium in a special case.

Example 2.5
The number of annual losses has a Poisson distribution with a mean of 5. The size of each loss
has a two-parameter Pareto distribution with θ = 10 and α = 2.5. An insurance for the losses
has an ordinary deductible of 5 per loss. Calculate the expected value of the aggregate annual
payments for this insurance.
Solution. Let X be the loss random variable. Then, (X − 5)+ , is the claim random variable.

10
E(X) = = 6.667
2.5 − 1
 "  2.5−1 #
10 10
E(X ∧ 5) = 1− = 3.038
2.5 − 1 5 + 10
E [(X − 5)+ ] = E(X) − E(X ∧ 5)
= 6.667 − 3.038 = 3.629

29
 Expected aggregate claims is given by

E(N)E [(X − 5)+ ] = 5(3.629) = 18.15

Theorem 2.1
Suppose that Pr(a < S < b) = 0. Then, for a ≤ d ≤ b, we have

b−d d −a
E[(S − d)+ ] = E[(S − a)+ ] + E[(S − b)+ ].
b−a b−a

Proof. Let a ≤ x ≤ b. Since a ≤ x and FS (x) is nondecreasing, we have FS (a) ≤ FS (x). On the other
hand, Z x Z Z a Z x b
FS (x) = fS (y)dy = fS (y)dy + fS (y)dy ≤ FS (a) + fS (y)dy = FS (a).
0 0 a a
Hence, FS (x) = FS (a) for all a ≤ x ≤ b. Next, we have;
Z ∞
E[(S − d)+ ] = [1 − FS (x)]dx
d
Z ∞ Z d
= [1 − FS (x)]dx − [1 − FS (x)]dx
a a
Z d
= E[(S − a)+ ] − [1 − FS (a)]dx
a
= E[(S − a)+ ] − (d − a)[1 − FS (a)].

The above is true for all a ≤ d ≤ b. In particular, if d = b, we have

E[(S − a)+ ] − E[(S − b)+ ]

E[(S − b)+ ] = E[(S − a)+ ] − (b − a)[1 − FS (a)] ⇒ 1 − FS (a) = .
b−a
Hence,
d −a
E[(S − b)+ ] = E[(S − a)+ ] − {E[(S − a)+ ] − E[(S − b)+ ]}
b−a
b−d d −a
= E[(S − a)+ ] + E[(S − b)+ ]
b−a b−a

Example 2.6
A reinsurer pays aggregate claim amounts in excess of d, and in return it receives a stop-loss
premium E[(S −d)+ ]. You are given E[(S −100)+ ] = 15, E[(S −120)+ ] = 10, and the probability
that the aggregate claim amounts are greater than 100 and less than 120 is 0. Calculate E[(S −
105)+ ].
Solution. We have,
   
120 − 105 105 − 100
E[(S − 105)+ ] = (15) + (10) = 13.75
120 − 100 120 − 100

More simplification result follows.

30
Theorem 2.2
Suppose S is discrete and Pr(S = kh) ≥ 0 for some fixed h and k = 0, 1, 2 · · · . Also, Pr(S = x) = 0
for all x ̸= kh. Then, provided d = jh, for any nonnegative integer j, we have
∞
E[(S − jh)+ ] = h ∑ {1 − FS [(n + 1) j]}.
n=0

In particular,
E[(S − ( j + 1)h)+ ] − E[(S − jh)+ ] = h[FS ( jh) − 1].

Proof. We have;

E[(S − jh)+ ] = ∑ (x − jh) fS (x)

x> jh
∞ ∞
= ∑ (kh − jh)Pr(S = kh) = h ∑ (k − j)Pr(S = kh)
k= j k= j
∞ k− j−1 ∞ k− j−1
=h∑ ∑ Pr(S = kh) − h ∑ ∑ Pr(S = kh)
k= j n=0 n=0 k=n+ j+1
∞
= h ∑ {1 − FS [(n + 1) j]}.
n=0

Finally, we have
∞
E[(S − ( j + 1)h)+ ] − E[(S − jh)+ ]s = h ∑ {1 − FS [(n + 1) j + n + 1]}
n=0
∞
− h ∑ {1 − FS [(n + 1) j]}
n=0
∞
= h ∑ {1 − FS [(n + 1) j]}
n=1
∞
− h ∑ {1 − FS [(n + 1) j]}
n=0
= h[FS ( jh) − 1]

Example 2.7
Given the following information about the distribution of a discrete aggregate loss random
variable:
x 0 25 50 75
FS (x) 0.05 0.065 0.08838 0.12306
Calculate E[(S−25)+ ], E[(S−50)+ ], E[(S−75)+ ], and E[(S−100)+ ], given that E(S) = 314.50.
Solution. We have;

E[(S − 25)+ ] = 314.50 − 25(1 − 0.05) = 290.75

E[(S − 50)+ ] = 290.75 − 25(1 − 0.065) = 267.375
E[(S − 75)+ = 267.375 − 25(1 − 0.08838) = 244.5845
E[(S − 100)+ ] = 244.5845 − 25(1 − 0.12306) = 222.661.

31
Example 2.8
Prescription drug losses, S, are modeled assuming the number of claims has a geometric
distribution with mean 4, and the amount of each prescription is 40. Calculate E[(S − 100)+ ].
Solution. Let N denote the number of prescriptions and S the aggregate losses. Then, S = 40N.
We have;

E[(S − 100)+ ] = E(S) − E(S ∧ 100)

= 40E(N) − 40E(N ∧ 2.5)
= 40E(N) − 40 [ fN (1) + 2 fN (2) + 2.5(1 − FN (2))]
= 40(4) − 40[0.16 + 2(0.1280) + 2.5(0.5120)]
= 92.16

4n
where fN (n) = .
5n+1

Example 2.9
For a collective risk model:

(i) The number of losses has a Poisson distribution with λ = 2.

(ii) The common distribution of the individual losses is:

x f (x)
1 0.6
2 0.4

An insurance covers aggregate losses subject to a deductible of 3. Calculate the expected

aggregate payments of the insurance.
Solution. We have

E[(S − 3)+ ] = E[(S − 2)+ ] + FS (2) − 1

= E[(S − 1)+ ] + FS (2) + FS (1) − 2
= E(S) + FS (0) + FS (1) + FS (2) − 3
= E(S) + 3 fS (0) + 2 fS (1) + fS (2) − 3
e−2 22
= 2(1.4) + 3e−2 + 2e−2 (0.6) + 2e−2 (0.4) + (0.6)2 − 3
2!
= 0.7365

Example 2.10
A compound Poisson claim distribution has λ = 5 and individual claim amounts distributed as
follows:
x f (x)
5 0.6
k 0.4
where k > 5. The expected cost of an aggregate stop-loss insurance subject to a deductible of 5
is 28.03. Calculate k.

32
Solution. The stop-loss insurance with deductible 5 pays

(S − 5)+ = S − (S ∧ 5).

Thus,
E[(S − 5)+ ] = E(S) − E(S ∧ 5).
We have
E(S) = E(N)E(X) = 5[5(0.6) + k(0.4)] = 15 + 2k
and

E(S ∧ 5) = 5[1 − FS (5)]

= 5[1 − Pr(S = 0)]
= 5[1 − Pr(N = 0)]
= 5(1 − e−5 ) = 4.9663.

Thus,
28.03 = E[(S − 5)+ ] = 15 + 2k − 4.9663 ⇒ k = 9.

Example 2.11
WidgetsRX owns two factories. It buys insurance to protect itself against major repair costs.
Profit equals revenue, less the sum of insurance premiums, retained major repair costs, and all
other expenses. WidgetsRX will pay a dividend equal to the profit, if it is positive.
You are given:

(i) Combined revenue for the two factories is 3.

(ii) Major repair costs at the factories are independent.

(iii) The distribution of major repair costs (k) for each factory is

k Probability(k)
0 0.4
1 0.3
2 0.2
3 0.1

(iv) At each factory, the insurance policy pays the major repair costs in excess of that factory’s
ordinary deductible of 1. The insurance premium is 110% of the expected claims.

(v) All other expenses are 15% of revenues.

Calculate the expected dividend.

Solution. Let R denote the retained major repair cost for both factories. That is, this is the portion
of the major repair costs to be covered by the ownership of the factory. Note that R is an aggregate
random variable and R = 0, 1, 2. Let D stand for the dividend. The expected claims per factory
for the insurance company is

E[(X − 1)+ ] = (2 − 1)(0.2) + (3 − 1)(0.1) = 0.4.

33
 Thus, the insurance company charges WidgetsRX a premium of (1.10)(2)(0.4) = 0.88. Hence,
the formula of the dividend can be expressed as

D = 3 − 0.88 − R − 0.15(3) = 1.67 − R.

The distribution of R is found as follows:

For R = 0, both factories have no major repair costs and in this case Pr(R = 0) = (0.4)(0.4) =
0.16. For R = 1, one factory has a major repair cost and the other does not. In this case, Pr(R =
1) = 2(0.4)(0.6) = 0.48. Hence, Pr(R = 2) = 1 − Pr(R = 0) − Pr(R = 1) = 0.36. The distribution
of D is
D Probability (D)
1.67 0.16
0.67 0.48
0 0.36
Finally,
E(D) = 1.67(0.16) + 0.67(0.48) = 0.58888.

2.1 EXERCISES

1. You are given: E[(S − 15)+ = 0.34 and E[(S − 30)+ ] = 0.55. Calculate FS (15).

2. An aggregate claim distribution has the following characteristics: Pr(S = i) = 16 for

i = 1, 2, 3, 4, 5, 6. A stop-loss insurance with deductible amount d has an expected insurance
payment of 1.5.
Find d.

3. You are given:

(i) S takes only positive integer values.

(ii) E(S) = 53 .
(iii) E[(S − 2)+ ] = 16 .
(iv) E[(S − 3)+ ] = 0.

Find the value of fS (1).

4. For a stop-loss insurance on a three person group:

(i) Loss amounts are independent.

(ii) The distribution of loss amount for each person is:
Loss Amount Probability
0 0.4
1 0.3
2 0.2
3 0.1
(iii) The stop-loss insurance has a deductible of 1 for the group.

Calculate the net stop-loss premium.

34
 1
5. Suppose that the aggregate loss random variable is discrete satisfying Pr(S = 50k) = 2k+1
for
k = 0, 1, 2, · · · and Pr(S = x) = 0 for all other x.

Calculate E[(S − 150)+ ].

6. For a certain company, losses follow a Poisson frequency distribution with mean 2 per year, and
the amount of a loss is 1, 2, or 3, each with probability 1/3. Loss amounts are independent of the
number of losses, and of each other.
An insurance policy covers all losses in a year, subject to an annual aggregate deductible of 2.

Calculate the expected claim payments for this insurance policy.

7. The number of annual losses has a Poisson distribution with a mean of 5. The size of each loss
has a two-parameter Pareto distribution with θ = 10 and α = 2.5. An insurance for the losses has
an ordinary deductible of 5 per loss.

Calculate the expected value of the aggregate annual payments for this insurance.

8. In a given week, the number of projects that require you to work overtime has a geometric
distribution with β = 2. For each project, the distribution of the number of overtime hours in the
week is the following:

x f (x)
5 0.2
10 0.3
20 0.5

The number of projects and number of overtime hours are independent. You will get paid for
overtime hours in excess of 15 hours in the week.

Calculate the expected number of overtime hours for which you will get paid in the week.

2.2 Individual Policy Modification Impact on Aggregate Losses

Any individual policy modifications such as deductibles, policy limits, or coinsurance will have an
impact on the aggregate loss/payment random variable. In this lecture, it will be assumed that the
ground-up distribution of the individual loss amount X is unaffected by policy modifications, and only
the payments themselves are affected.

We next start by reminding the reader of the following notations: The number of losses will be denoted
by NL ; the number of payments by NP , the probability of a loss resulting in a payment by v; the amount
of payment on a per-loss basis is XL with XL = 0 if a loss results in no payment; and the amount of
payment on a per-payment basis is XP where only losses that result on a non-zero payment are
considered and the losses that result in no payment are completely ignored. Note that Pr(XP = 0) = 0
and XP = XL |XL > 0.

The cdfs of XL and XP are related as follows:

FXL (y) = Pr(XL ≤ y|XL = 0)Pr(XL = 0) + Pr(XL ≤ y|XL > 0)Pr(XL > 0)
= (1 − v) + vFXP (y).

35
The mgfs of XL and XP are related as follows:

MXL (t) = E(etXL |XL = 0)Pr(XL = 0) + E(etXL |XL > 0)Pr(XL > 0)
= (1 − v) + vMXP (t).

The pgfs of NP and NL are related as follows (Recall from Actuarial Risk Theory):

PNP (z) = PNL (1 − v + vz).

Now back to the aggregate payments. On a per-loss basis, the total payments may be expressed as

S = XL1 + XL2 + · · · + XLNL

With S = 0 if NL = 0 and where XL j is the payment amount on the jth loss. Alternatively. ignoring all
losses that do not result in a non zero-payment, the aggregate S can be expressed as

S = XP1 + XP2 + · · · + XPNP

with S = 0 if NP = 0 and where XPj is the payment amount on the jth loss which results in a non-zero
payment. On a per-loss basis, S is a compound distribution with primary distribution NL and secondary
distribution XL so that
MS (t) = PNL [MXL (t)] .
Likewise, on a per-payment basis, we have

MS (t) = PNP [MXP (t)] .

Note that
PNL [MXL (t)] = PNL [1 − v + vMXP (t)] = PNP [MXP (t)] .
Note that individual policy modifications factor either in the expression of XL or XP and thus have an
impact on the aggregate payment.

Example 2.12
A ground-up model of individual losses follows a Pareto distribution with α = 4 and θ = 10.
The number of losses is a Poisson distribution with λ = 3. There is an ordinary deductible of 6,
a policy limit of 18 – applied before the ..... and coinsurance of75%. Find the expected value and
the variance of S, the aggregate payments on a per-payment basis.

Example 2.13
A company insures a fleet of vehicles. Aggregate losses have a compound Poisson distribution.
The expected number of losses is 20. Loss amounts, regardless of vehicle type, have exponential
distribution with θ = 200. In order to reduce the cost of the insurance, two modifications are to
be made:

(i) a certain type of vehicle will not be insured. It is estimated that this will reduce loss
frequency by 20%.

(ii) a deductible of 100 per loss will be imposed.

Calculate the expected aggregate amount paid by the insurer after the modifications.
Solution. We want

E(S) = E(NL )E(XL ) = E(NL )E[(X − 100)+ ] = E(NL )[E(X) − E(X ∧ 100)].

36
 From Table C,
100
E(X ∧ 100) = 200(1 − e− 200 ).
Thus, h i
100
E(S) = (20)(0.8) 200 − 200(1 − e− 200 ) ≈ 1941.

2.2 EXERCISES

1. The number of ground-up losses in Poisson distribution with mean λ = 3. The individual loss
distribution is Pareto with parameter α = 4 and θ = 10. An individual ordinary deductible of
6, coinsurance of 75%, and an individual loss limit of 24 (before application of the deductible
and coinsurance) are all applied. Determine the mean, variance and distribution of aggregate
payments.

2. A ground-up model of individual losses has the gamma distribution with parameters α = 2 and
θ = 100. The number of losses has the negative binomial distribution with r = 2 and β = 1.5. An
ordinary deductible of 50 and a loss limit of 175 (before imposition of the deductible) are applied
to each individual loss.

(a) Determine the mean and variance of the aggregate payments on a per-loss basis.
(b) Determine the distribution of the number of payments
(c) Determine the cumulative distribution function of the amount XP of a payment given that a
payment is made.

37