Chapter 8 :

Frequency and severity with coverage modifications

8.2 Deductibles

Definition 8.1 Let d ≥ 0 such that P r(X > d) > 0.

• The per-payment

variable (excess loss) is

, X ≤ d;




P

Y =
, X > d.






• The per-loss variable is 


, X ≤ d;




L

Y = = =
, X > d.






• d is called (ordinary) and Y P = .

• P (Y L = 0) = .

• E(Y P ) = .

SY P (y) = P r(Y P > y) =

FY P (y) = 1 − SY P (y) =

fY P (y) =

hY P (y) =

1
FY L (y) = P r(Y L ≤ y) =

SY L (y) = 1 − FY L (y) =


, y > 0;





fY L (y) = 
, y = 0.







, y > 0;





hY L (y) = 
, y = 0.





Definition 8.2 A deductible modifies the ordinary de-

ductible by adding the deductible when there is a positive amount
paid.


, X ≤ d;




Y∗L

• The per-loss variable is =
, X > d.
 




• The per-payment variable

 is

, X ≤ d;




Y∗P =

=
, X > d.
 






, 0 ≤ y ≤ d;




FY∗L (y) = P r(Y∗L ≤ y) = 


, y > d.
 





, 0 ≤ y ≤ d;





SY∗L (y) = 1 − FY∗L (y) = 
, y > d.
 




2
 

, y = 0;











fY∗L (y) = 

, 0 < y < d;




, y > d.







fY L (y) , 0 < y < d;





hY∗L (y) = ∗ =
SY∗L (y) 


 , y > d.
P
S
 Y ∗
P (y) = P r(Y∗ > y) = =

, 0 ≤ y ≤ d;






, y > d.







, 0 ≤ y ≤ d;











FY∗P (y) = 1−SY∗P (y) = 
 



, y > d.







, 0 < y < d;










fY∗P (y) = FY′ ∗P (y) = 







, y > d.






, 0 < y < d;






fY∗P (y)





hY∗P (y) = =
SY∗P (y) 



, y > d.






3
Theorem 8.3: The expected cost
per loss per payment
L P E(X) − E(X ∧ d) E(Y L)
E(Y ) = E(X) − E(X ∧ d) E(Y ) = =
SX (d) SX (d)
L P E(X) − E(X ∧ d) E(Y∗L)
E(Y∗ ) = E(X) − E(X ∧ d) + dSX (d) E(Y∗ ) = +d=
SX (d) SX (d)
Recall that E[(Y L)k ] = and
E(X) = . Then
E[Y L] = and
E[Y P ] = E[Y L |Y L > 0] =


, X≤d




Since Y∗L − Y L = 


, X > d,
 




we have E(Y∗L ) =
Also,

E(Y∗P ) = E(Y∗L |Y∗L > 0) =

8.3 The loss elimination ratio and the effect of inflation

for ordinary deductibles

Definition 8.4 The (LER) is the ratio of the

decrease in the expected payment with an ordinary deductible to the
expected payment without the deductible.
LER=
4
Theorem 8.5: For an ordinary deductible of d after uniform infla-
tion of 1 + r, the expected cost per loss is

E[(Y − d)+] =

where Y = (1 + r)X. If > 0, then the expected cost per

payment is

Proof: SY (y) =

E[Y ∧ d] =

⇒ E[Z L ] = , and
by Theorem 8.3, E[Z P ] =

8.4 Policy limits



, X < u;





The limited loss variable is Y = =
, X ≥ u.








, y < u;





FY (y) = P r(Y ≤ y) = 
, y ≥ u.
 





, y < u;





SY (y) = 1 − FY (y) = 
, y ≥ u.





5
 

, y < u;





fY (y) = 
, y = u.
 




hY (y) = .

Note that (X − d)+ + (X ∧ d) =

⇒ E(X) =

Theorem 8.6: For a policy limit of u, after uniform inflation of

1 + r, the expected cost per loss is E[Y ∧ u] =
where Y = (1 + r)X.

8.5 Coinsurance, deductibles, and limits

Coinsurance only: Y = αX, 0 ≤ α ≤ 1.

inflation → policy limit → deductible → coinsurance:

The per loss variable Y L = .



,










YL =


 
,



,






where d∗ = d/(1 + r) and u∗ = u/(1 + r).

In this definition, the policy limit is , the maximum amount

6
payable; and u is the loss above which no additional benefits are paid
and is called the .

Note that Y L = .

The per payment variable Y P = Y L|Y L > 0.



,










YP =


 
,



,






Theorem 8.7: E[Y L ] =

and E[Y P ] = .

Proof: Y L =

E[Y P ] = E[Y L |Y L > 0] =

Theorem 8.8:
E[(Y L)2] = α2 (1+r)2{E[(X ∧u∗)2]−E[(X ∧d∗)2]−2d∗[E(X ∧u∗)−E(X ∧d∗)]}
P 2 E[(Y L )2]
and E[(Y ) ] = .
SX [d∗]

Proof: (Y L ) =

7
We declare that

because both sides equal

Making this substitution and taking expectation on both sides gives

the proof.
Example: (Q.7, Fall, 2004, SOA Exam M/3) Annual prescription drug costs X ∼
Pareto (α, θ) with α = 2 and θ = 2000. A prescription drug plan pays annual drug costs for
an insured member subject to the following provisions:
(1) The insured pays 100% of costs up to the ordinary annual deductible of 250.
(2) The insured then pays 25% of the costs between 250 and 2250.
(3) The insured pays 100% of the costs above 2250 until the insured has paid 3600 in total.
(4) The insured then pays 5% of the remaining costs.
Determine the expected annual plan payment.

The insured
 pays

, X < d1 ;











, d1 ≤ X < d2 ;




L

Y =
, d2 ≤ X < d3 ;
 









, d3 ≤ X.






d1 = , α1 = , d2 = , α2 = , α3 =
and d3 = by

E[Y L] =

8
is easy to compute.

E[Y L] =

8.6 The impact of deductibles on claim frequency

Let M1, M2, . . . be identically and independently distributed random

variables with pgf PM (z), N be a counting random variable with pgf
PN (Z). Assume that N is independent of Mj s. Then the pgf of

9
S = M1 + M2 + . . . + MN (where S = if N = 0) is PS (z) =
where PN (z) and PM (z) are called the
and distributions, respectively. If N follows the
Poisson/Binomial/Negative Binomial distribution, then S is called
Poisson/Binomial/Negative Binomial distribution.

∗n
PM (z) =

PS (z) =

Theorem 7.4: Suppose the pgf PN (z; θ) satisfies PN (z; θ) =

for some parameter θ and some function B(z) that is independent
of θ. There may be other parameters as well, and they may appear
anywhere in the pgf. Then PS (z) = PN [PM (z); θ] can be rewritten
as PS (z) =

where f0 = , cM = and cT = .

• B(m, θ): PN (z; θ) = ⇒ B(z) = ;

• NB(r, θ): PN (z; θ) = ⇒ B(z) = ;

• Poisson(θ): PN (z; θ) = ⇒ B(z) = .

10
 M
Proof: Since PM (z) =

⇒ PM (z) =

where

Therefore PM (z) =

⇒ PS (z) = PN [PM (z); θ] =

This shows that deleting or modifying is equivalent

to modifying the parameter θ of the primary distribution PN (z; θ)
because it does not add a new distribution.

Let d be the deductible,

v = Pr(a loss results in a payment)= and



, the ith loss result in a payment;







Ij = 
, otherwise.






Then Ij ∼ and PIi (z) = .

Let N L be the number of losses. If I1, I2, · · · are mutually indepen-

dent and are also independent of N L , then N P =
is the number of payments, and has a compound distribution with

11
PN P (z) = .

If the pgf of N L depends on a parameter θ such that PN L (z) =

PN L (z; θ) = where B(z) is functionally independent of
θ (see Theorem 7.4), then PN P (z; θ) =

implying N L and N P are from the parametric distribution

family and only the parameter need be changed.

Recall that P M (z) =

where cM = .

Suppose the pgf of N L depends on a parameter θ and α such that

PN L (z) = PN L (z; θ, α) =

where α = . If is a pgf, then PN L (z) =

is the pgf of a distribution, P (z) = with
p0 = , and pM
0 = .

PN P (z; θ, α) =

12
⇒ N P and N L are from the parametric distribution family,
where θ∗ = and

α∗ =

If N L is zero-truncated, that is, , then

α∗ = ,

implying N P is .

ZM Poisson: If N L ∼ Poisson(λ), that is, PNPoisson

L (z) = eλ(z−1) ,

PNZM
L
Poisson(z) =

⇒ B(z) = ,α= and θ = .

PNZM
P
Poisson(z; α, λ) = P ZM
NL
Poisson(z; α∗, λ∗)

⇒ λ∗ = and

(pM ∗ ∗
0 ) = α =

ZM Binomial: If N L ∼ Binomial(m, q), that is, PNBinomial

L (z) =
[1 + q(z − 1)]m, then

PNZM
L
Binomial(z) =

13
⇒ B(z) = ,α= and θ = .

PNZM
P
Binomial(z; α, m, q) = P ZM
NL
Binomial(z; α∗, m∗, q ∗)

⇒ m∗ = , q∗ =

(pM ∗ ∗
0 ) = α =

ZM NB: If N L ∼ NB(r, β), that is, PNNB

L (z) = [1 − β(z − 1)]
−r

(r 6= 0), then

PNZM
L
NB (z) =

⇒ B(z) = ,α= and θ = .

PNZM
P
NB (z; α, r, β) = P ZM
NL
NB (z; α∗, r∗, β ∗ )

⇒ r∗ = , β∗ = and

(pM ∗ ∗
0 ) = α =

ZM logarithmic: If N L ∼ logarithmic(β), that is,

logarithmic ln(1 + β) − ln[1 − β(z − 1)]
PN L (z) =
ln(1 + β)
logarithmic
with PN L (0) = 0 (see Section 6.6), then
14
 ZM logarithmic
PN L (z) =

⇒ B(z) = ,α= and θ = .

logarithmic logarithmic
PN P (z; α, β) = PN L (z; α∗ , β ∗)

⇒ β∗ = and

(pM ∗ ∗
0 ) = α =

Table 8.3 Frequency adjustments

NL parameters for N P
ZM Poisson(λ)• θ = λ, λ∗ = vλ, B(z) = ez , and PN L (z) = eλ(z−1)
ZM Binomial(m, q)• θ = q, m∗ = m, q ∗ = vq, B(z) = (1 + z)m , and PN L (z) = [1 + q(z − 1)]m
ZM NB(r, β)• θ = β, r ∗ = r, β ∗ = vβ, B(z) = (1 − z)−r , and PN L (z) = [1 − β(z − 1)]−r
ZM Logarithmic(β) θ = β, β ∗ = vβ, B(z) = 1 + ln(1 − z), and PN L (z) = ln(1+β)− ln[1−β(z−1)]
ln(1+β)
• PN L (0) = B(−θ).
P (z) − p0
P M (z) = pM M T M M
0 + (1 − p0 )P (z) = p0 + (1 − p0 ) where p0 = P (0)
1 − p0
B[θ(z − 1)] − B(−θ)
PN L (z; θ, α) = α + (1 − α) where α = pM L
0 = P r(N = 0) = PN L (0)
1 − B(−θ)
B[vθ(z − 1)] − B(−vθ)
PN P (z; θ, α) = α∗ + (1 − α∗ ) = PN L (z; v ∗ , α∗ ) where v ∗ = v θ
1 − B(−vθ)
B(−vθ) − B(−θ)
α∗ = (pM P
0 ) = P r[N = 0] = PN P (0) = PN L (1 − v; θ, α) = α + (1 − α)
∗
1 − B(−θ)

15
Sometimes, we want to determine the distribution of N L from that
of N P . For example, data may have been collected on the number
of payments in the presence of a deductible and from that data the
parameters of N P can be estimated. We may want to know the
distribution of payments if the deductible is removed.

Since PN P (z) = , let w = . Then

z= and PN L (w) = , implying the
formulas derived perviously hold with v replaced by . Note that
it is possible that the resulting pgf for N L valid (for example,
negative probabilities, see example of textbook).

If N L has a compound distribution, then we can write PN L (z) =

and therefore PN P (z) = ⇒
N P also has a distribution.

For N L ∼ compound Poisson, we have PN L (z) = and

PN P (z) = .
∗
Let N d and N d be the frequency random variables when the de-
ductibles are d and d∗,

respectively, and v =

16
Then Table 8.3 can be used to move from the parameters of N d to
∗
the parameters of N d .

If d∗ > d, we have and the formulas will lead to a legitimate

∗
distribution for N d .

If d∗ < d, we have and there is no assurance that a legitimate

distribution will result.

9.7 The impact of individual policy modifications on ag-

gregate payments

Recall Y P = Y L |Y L > 0 and v = P r(Y L > 0) = P r(X > d)) =

SX (d). Since FY L (y) = and

FY P (y) =

we have FY L (y) = . Also,

MY L (t) =

On a per loss basis, the total payments may be expressed as S =

with S = 0 if where YjL ≥ 0 is the
payment amount on the j th loss.

On a per payment basis, the total payments may be expressed as

17
S = with S = 0 if where YjP > 0 is the
payment amount on the j th loss, which results in a nonzero payment.

The moment generating function of S on a per-loos basis is

MS (t) = E[etS ] =

which is a moment generating function. Similarly, E[etS ] =

. So,

PN P [MY P (t)] =

versus PN P [z] =

18