You are on page 1of 9

RESEARCH ARTICLE | JUNE 01 2020

Premium determination with multidimensional Bühlmann-


Straub credibility model 
M. R. S. Aji; S. Nurrohmah  ; I. Fithriani

AIP Conf. Proc. 2242, 030026 (2020)


https://doi.org/10.1063/5.0007954

CrossMark

 
View Export
Online Citation

26 December 2023 03:18:49


Premium Determination with Multidimensional
Bühlmann-Straub Credibility Model
M. R. S. Aji, S. Nurrohmaha) and I. Fithriani

Department of Mathematics, Faculty of Mathematics and Natural Sciences (FMIPA), Universitas Indonesia,
Depok 16424, Indonesia
a)
Corresponding author: snurrohmah@sci.ui.ac.id

Abstract. Every insurance company must certainly be able to fulfill its obligations as a guarantor if there are claims
submitted by policy holders or insured in the future. Therefore, premium determination is very important for insurance
companies. The premium determined by the insurance company contains a net-premium which is an estimate of claims
that will be made during the insurance period. So that the net-premium can be determined properly, the insurance company
can use the observations of claims that have been made by policy holders or a risk class by considering claims observations
from other policy holders or other risk classes that still have similar risks. One of model that can be used in net-premium
calculations is the Bühlmann-Straub credibility model. The more information about claims made by policy holders, the

26 December 2023 03:18:49


net-premium estimation produced is also getting better, because larger observation objects will provide higher accuracy.
In some cases, there are insurance products with claims that are relatively rare, therefore if possible the insurance companies
can exchange information either directly or through the association of insurance company. However, the Bühlmann-Straub
credibility model only considers the information held by the insurance company itself, so in this paper the multidimensional
Bühlmann-Straub credibility model is used to determine premium, so that the net-premium calculations can be done using
observations from sources of information or insurance company. The Multidimensional Bühlmann-Straub credibility
model can be formed by generalizing from the one-dimensional Bühlmann-Straub credibility model. The multidimensional
Bühlmann-Straub credibility model consists of three parameters which must be estimated, namely , , and , where these
parameters can be estimated nonparametrically.

Keywords: Bühlmann-Straub credibility theory, multidimensional credibility model, net premium

INTRODUCTION
Risk can be defined as the potential loss due to a certain event [1]. Moving part or all of the losses that might occur
due to a risk is a very good thing to do. Transfer of losses due to these risks can be done to insurance companies with
buy an insurance product to insure the object you want to guarantee.
Insurance is an agreement between two parties, namely an insurance company and policy holder [2]. The policy
holder will pay a sum of money called a premium to the insurance company and instead, the insurance company will
provide coverage to policy holders for future losses in accordance with the provisions in the policy. In order to be able
to fulfill expenses as a form of insurance in the future, then premium determination is an important thing for insurance
companies.
Credibility theory is a tool that can be used in determining the amount of net premiums based on observations of
claims from risk class (individual claims history) by considering observations of claims from another risk class who
still have similar risks (group claim history). It is an experience rating technique to determine insurance premiums by
combining two fundamental concepts, the individual risk and the collective (or industry) risk [3].

Proceedings of the 5th International Symposium on Current Progress in Mathematics and Sciences (ISCPMS2019)
AIP Conf. Proc. 2242, 030026-1–030026-8; https://doi.org/10.1063/5.0007954
Published by AIP Publishing. 978-0-7354-2001-4/$30.00

030026-1
Two of the credibility models that often used are Bühlmann credibility model and Bühlmann-Straub credibility
model. Bühlmann developed a linear credibility formula under the principle of minimum mean square error (least
squares). Without assuming a prior distribution with specified distributional parameters, Bühlmann considered the
best linear estimator based on observed claims, which can be estimated consistently by the method of moments. The
Bühlmann-Straub credibility model is a development of Bühlmann credibility model [4].
The Bühlmann-Straub credibility model mentioned earlier is one-dimensional, meaning that it only uses one
information of claim amount made by risk class number i in the period number j. In some cases, there are insurance
products with claims that are relatively rare, so the insurance companies that want to take into account claims need
additional information besides those owned by the insurance company. Additional information can be obtained from
the association of the insurance company or other insurance companies so that a better premium can be obtained. To
be able to determine the amount of premium by considering other information, in this paper we will consider the
multidimensional Bühlmann-Straub credibility model which is a generalization of the Bühlmann-Straub one-
dimensional credibility model.

METHOD

Bühlmann-Straub Credibility Model

In this section we will discuss the Bühlmann-Straub credibility model. Suppose there are , … , with state
the average of claim amount from policy holders in period j. Therefore unknown state the risk parameters that
explain the risk characteristics, state average of the claim amount with a given value of (called the hypothetical

average of the process variance, state variance of the hypothetical mean, = +


mean), state variance of the claim amount with a given value of (called the process variance), state average
of the hypothetical mean,

⋯+ =

26 December 2023 03:18:49


, and . Based on the claims of each period, an estimate of claims amount for one future period for
each risk class denoted by , where states the estimated amount of claims for the next period that will be
carried out, can be written as [3]:

= + 1− (1)

=
"
where,
" $ %⁄&
(2)

General Form of Multidimensional Credibility Model

In this section we will discuss the general form of multidimensional credibility model that will be the basis for
establishing a multidimensional Bühlmann-Straub credibility model. Suppose there are risk classes as much as r where

, , … , ( , then the second risk class there are p information


for each risk class there is an observation of claims amount from every insurance companies, so for the 1st risk class

, ,…,
there are p information sources consisting of
(

consisting of ) , ) , … , ) ( with * + state the amount of claims made by the risk class i to the insurance
sources consisting of and so on up to the risk class number r there are p information sources

company-k for , = 1,2, … , . Therefore unknown * state the risk parameters that explain the risk characteristics for

Based on the claims of each risk class that is . , . , … , .) , where .* / = 0 * , * , … , * ( 1, an estimate of


risk class i. Another notation which will be used is explained in Table 1.

2 for 3 = 1,2, … , 4, where =


/
2
5 , , … , 6
claims amount for one future period for each risk class denoted by
2 2 ( 2 [5] with 2 states the estimated amount of claims for the next period that will be
carried out by the risk class i at the first insurance company, and so on up to ( 2 states the estimated amount of
claims for the next period that will be carried out by the risk class i on the insurance company number p.

030026-2
TABLE 1. Notation and the interpretation.
Notation Interpretation
+ * The average of the claim amount of risk class number i to the insurance
company number k with a given value of * , is called the hypothetical mean
of risk class number i and insurance company number k
+ * The variance of the claim amount of risk class number i to the insurance
company number k with a given value of * , is called the process variance of

78 9=
risk class number i and insurance company number k
+ * + The average of the hypothetical mean of each risk class in the insurance

9=
company number k
78 + * + The average of the process variance of each risk class in the insurance
company number k
: 40μ+ * 1 = <+ The variance of the hypothetical mean of each risk class in the insurance
company number k
=>:0 ? * , @ * 1 = <?@ The covariance of the hypothetical mean of each risk class in the insurance
company number c and d

For each risk class i, given a vector .* / = 0 * , * , … , * ( 1 which meets the properties [6]:
• Observations * A * , * A * , … , * ( A * are independent, in other words observations from different

• Observations * A * , * A * , … , * ( A * have a mean,


insurance companies for the risk class number i do not affect each other.

26 December 2023 03:18:49


70 * A *1 = * , 70 * A *1 = * , … , 70 *
(
A *1 = ( *

or in vector form can be written as follows,

70 * A *1 *
⎛70 A * 1⎞ = J
7 .* | =⎜ *

*
K=
*
⋮ ⋮ * (3)

⎝70 A * 1⎠
( ( *
*

with a covariance matrix of size × as follows,

* ⋯ 0
M * = N>:0.* , .* O A * 1 = P ⋮ ⋱ ⋮ S
0 ⋯
(4)
( *

• , ,…, ) independent and identically distributed.

There are three parameters that must be estimated in the general form of the multidimensional credibility model,
the three parameters are defined as follows;

78 * 9
⎛ 78 9⎞
=J ⋮ K= *
= 78 9
⋮ * (5)
( ⎝7T ( * U⎠

030026-3
⋯ 0
V = 78M * 9=P ⋮ ⋱ ⋮ S
0 ⋯
(6)
(

< < ⋯ <(


⎛< < ⋯ < (⎞
= N>: , =⎜
⋮ ⎟
O
⋮ ⋮ ⋱
* * (7)
<( <( ⋯ <(
⎝ ⎠

For each risk class, an estimation of the vectors * =0 * , * ,…, ( * 1 by estimating the entries of

number i for each insurance company that is * , * , … , * ( . The estimator is denoted by =


the vector using linear combinations of observations of the claims amount that have been made by the risk class
/
2
5 2 , 2 , … , ( 2 6 can be obtained by the following equation,

2 = W* ∙ .* + Y − W* ∙ (8)

with credibility factor for risk class number i,

W* = +V Z
(9)

RESULTS AND DISCUSSION

26 December 2023 03:18:49


Multidimensional Bühlmann-Straub Credibility Model
In this section we will establish Multidimensional Bühlmann-Straub credibility model using the general form that
has been obtained in the previous section. Suppose there as much as r risk class where for each risk class there is an

period there are p information sources consisting of * , * , … , * ( , then in second period there are p
observation of claims amount from every insurance companies in n period, so for the risk class number i in the first

information sources consisting of * , * , … , * ( and so on up to period number n there are p information


sources consisting of * , * , … , * ( . * + states the average amount of claims made by as much as * +
policy holders in the risk class number i to the insurance company number k in the period number j, where * + states
number of policy holders of the risk class number i, insurance company number k in the period number j.

For each risk class i, given a vector .* / = 0 * , * , … , * ( 1 which meets the properties:
The assumptions used for the general form of the multidimensional credibility model can be written as follows;

• Observations * A * , * A * , … , * ( A * independent. In other words, the observations from


different insurance companies for the risk class number i and periode number j did not affect each other.

• Observations * A *, * A *, … , *
(
A * have a mean,

70 * A *1 = * , 70 * A *1 = * , … , 70 *
(
A *1 = ( *

or in vector form can be written as follows,

70 * A *1 *
⎛70 A * 1⎞ = J
70.* A * 1 = ⎜ *

*
K=
⋮ ⋮ * (10)

⎝70 A * 1⎠
( ( *
*

030026-4
with a covariance matrix × as follows,

⋯ 0
% [ \]

⎛ "]

N>:0.* , .* O
A *1 = ⎜ ⋮ ⋱ ⋮ ⎟ (11)
0 ⋯
%^ [ \]

⎝ "] ^

• , ,…, ) mutually independent and identical distributed.

Then diagonal matrix _* and _*∙ are formed as follows,

* ⋯ 0 * ⋯ 0
_* = P ⋮ ⋱ ⋮ S , _* = P ⋮ ⋱ ⋮ S
0 ⋯ 0 ⋯
(12)
( (
* *

=∑
where,
+ +
* ` * . (13)

The observation that are known and used in general form of a multidimensional credibility model is just one

periods, so the multidimensional Bühlmann-Straub credibility model use vectors a* as a representation of n


observation period, whereas in the multidimensional Bühlmann-Straub credibility model observations are known in n

observation periods that is .* , .* , … , .* , which can be written as follows,

b* = 0c* c* … c* 1
( /

26 December 2023 03:18:49


(14)

where,
c* =∑
e
+ d "]
` * "] f (15)

* will be carried out by estimating the entries of the vector


using linear combinations of observations of claims that have occurred, that is c* , c* , … , c* ( . The estimator is
For each risk class, an estimation of the vector

denoted by 2
g
=5 2 , 2 ,…, ( 2 6 can be obtained by the following equation,

h 2 = W* ∙ b* + Y − W* ∙ (16)

with credibility factor,

W* = ∙ + _* Z[ ∙ V ∙ _* Z[ Z
(17)

Estimation of the Parameters

Estimating the , V and vectors is done by estimating the entries of the vector. Estimating the vector entries
+, + , <+ and <+i for , = 1,2, . . , and k = 1,2, . . , will be done nonparametrically with the following formula
[4, 5],

l+ = ∑)*` *
+
c* +
" f
(18)

where +
= ∑)*` *
+
,

030026-5
m
+ = ∑*`
)
∑ *
+
0 *
+
− c* +
1
) Z ` (19)

<m r
+ = maks <+ , 0

<r
)∙%m
+ == ∙ st − u
[
+ + f
" f

Z
= = s∑)*` 51 − 6 u
+ )Z "] f "] f
) " f " f
(20)

t = ∑)*` 0c* −c 1
+ ) + + "] f
)Z " f

c = ∑)*` c*
f
+ + "]
" f

<l
+i = sgn 5 6 ∙ min } ; •<m m
+ <i €
∗ ∗∗ ∗ ∗∗
xz
fy $xz
fy Axz
fy $xz
fy A

<z
+i = = ∑)*` 0c* −c 10c* −c i 1
∗ + ) + + i "] f
)Z " f
(21)

<z == ∑)*` 0c* −c 10c* −c i 1


∗∗ i ) + + i "] y
+i )Z " y

26 December 2023 03:18:49


Premium Calculation Using Multidimensional Bühlmann-Straub Credibility Model

The data used in this paper is sample data from the general insurance industry with property insurance products.
The data to be used is the total amount of claims made by policy holders, along with the number of policy holders
with property locations in Java which are differentiated based on their respective provinces that is DKI Jakarta, Central
Java, DI Yogyakarta, West Java, East Java and Banten that will be defined as risk classes {1,2,3,4,5,6} sequentially,
because inter-provinces are considered to have different conditions of vulnerability to disasters. The observation
period of the data used is as much as 4 years, that is from 2014 to 2017. For the purposes of calculating premiums
using the multidimensional Bühlmann-Straub credibility model, the data is divided into 3 parts with each part
representing a general insurance company with the product property insurance, so that data obtained for each general

W* and estimated amount of claim vector 2 for 3 = 1,2, … ,6.


insurance company A (Table 2), B (Table 3) and C (Table 4). Table 5 is the calculation result of credibility matrices

for the first insurance company or insurance company A is 514,084, while for the second insurance company or
For example, premium stimulation in risk class 1 or property in the DKI Jakarta province for the 5th or 2018 period

insurance company B is 524,079, and for the third insurance company or insurance company C is 705,274.

.‡ˆ Š‡ˆ ‰
TABLE 2. Data for insurance company A or insurance company -1.

Risk
Class-i j=1 j=2 j=3 j=4 j=1 j=2 j=3 j=4
1 878,742 494,894 405,738 382,930 231 221 199 276
2 321,429 1,933,965 1,867,775 1,069,744 42 43 47 110
3 1,400,000 3,971,220 138,471 1,419,499 10 9 14 30
4 2,500,000 2,666,667 974,423 1,189,572 3 6 3 7
5 849,300 3,805,623 2,245,930 2,153,573 40 37 49 81
6 688,557 602,488 420,828 508,943 21 24 32 84

030026-6
.‡ˆ Š‡ˆ ‹
TABLE 3. Data for insurance company B or insurance company -2.

Risk
Class-i j=1 j=2 j=3 j=4 j=1 j=2 j=3 j=4
1 644,579 710,653 345,415 261,860 231 221 199 276
2 291,927 1,583,488 2,265,462 843,629 42 43 47 110
3 1,400,000 2,116,927 250,000 2,139,022 10 9 14 30
4 2,500,000 2,528,000 974,423 1,428,571 3 6 3 7
5 207,958 1,575,095 1,438,144 3,075,695 40 37 49 81
6 1,126,700 1,571,294 420,828 402,854 21 24 32 84

.‡ˆ Š‡ˆ Œ
TABLE 4. Data for insurance company C or insurance company -3.
Œ
Risk
Class-i j=1 j=2 j=3 j=4 j=1 j=2 j=3 j=4
1 329,602 361,323 1,315,305 552,171 231 221 199 276
2 291,927 586,201 2,709,810 793,128 42 43 47 110
3 1,400,000 4,376,639 1,214,227 1,728,506 10 9 14 30
4 2,500,000 970,922 974,423 2,361,211 3 6 3 7
5 394,898 2,146,860 918,641 745,509 40 37 49 81
6 1,532,024 360,478 875,000 442,946 21 24 32 84

26 December 2023 03:18:49


TABLE 5. Credibility matrices W* and vectors 2 .

Credibility matrix W‡ •Ž
Risk

0.62356 0.40538 0.11083 514,084


Class-i

•0.49464 0.32157 0.08792 ‘ •524,079‘


0.13797 0.08969 0.02452 705,274
1

0.57423 0.37330 0.10206 1,248,250


• 0.45551 0.29613 0.08096‘ •1,106,462‘
0.12705 0.08260 0.02258 867,711
2

0.44029 0.28623 0.07825 1,475,840


•0.34926 0.22705 0.06208‘ •1,287,000‘
0.09742 0.06333 0.01731 918,067
3

0.25446 0.16542 0.04523 1,357,619


•0.20185 0.13122 0.03588‘ •1,193,220‘
0.05630 0.03660 0.01001 891,910
4

0.56401 0.36667 0.10024 2,042,646


•0.44741 0.29086 0.07952‘ •1,736,624‘
0.12479 0.08113 0.02218 1,043,475
5

0.54487 0.35422 0.09684 631,605


•0.43223 0.28099 0.07682‘ •617,304‘
0.12056 0.07837 0.02143 731,276
6

030026-7
CONCLUSION

The Multidimensional Bühlmann-Straub credibility model can be formed by generalizing from the one-
dimensional Bühlmann-Straub credibility model, so that premium calculations can be made taking into account data

model consists of three parameters which must be estimated, namely , V and , where these parameters can be
from several sources of information or insurance companies. Then the multidimensional Bühlmann-Straub credibility

estimated using nonparametric methods.

REFERENCES

1. Otoritas Jasa Keuangan, Peraturan Otoritas Jasa Keuangan Nomor 18/POJK. 03/2016 Tentang Penerapan
Manajemen Risiko Bagi Bank Umum (Indonesia, 2016).
2. Presiden Republik Indonesia, Undang-undang Republik Indonesia Nomor 40 Tahun 2014 Tentang Usaha
Perasuransian (Jakarta, 2014).
3. S. A. Klugman, H. H. Panjer and G. E. Willmot, Loss Models: From Data to Decision, 4th ed (Wiley, New York,
2012).
4. G. Pitselis, Insur. Math. Econ. 70, 373-386 (2016).
5. V. Pacáková, E. Šoltés and B. Linda, Ekon. Manag. 17, 170-184. (2014)
6. H. Bühlmann and A. Gisler, A Course in Credibility Theory and its Applications (Springer, Berlin, 2005).

26 December 2023 03:18:49

030026-8

You might also like