DETERMINING THE OPTIMAL PREMIUM FOR “BODA BODA” INSURANCE

USING CREDIBILITY MODEL

A project submitted to Kirinyaga University in Partial Fulfilment of the Requirement for a

Bachelor’s Degree in Actuarial Science.

Submitted by;

LUCAS MBEKE PA102/G/9349/20

SHIKANGA PRAISEHOPE MATEVA PA102/G/9377/20
MUGO DOUGLAS KAMAU PA102/G/8971/20
MURIAGORO JACKSON KAMAU PA102/G/8981/20
Declaration
Declaration by the Students
This project is our original work and all materials which are not our own have been cited and
acknowledged. The project has not been presented in this university or any other university for
the award of degree.

Group Member Registration Number Signed Date

LUCAS MBEKE PA102/G/9349/20 ….….. …………..
SHIKANGA PRAISEHOPE MATEVA PA102/G/9377/20 ……… ……….
MUGO DOUGLAS KAMAU PA102/G/8971/20 ……… …………..
MURIAGORO JACKSON KAMAU PA102/G/8981/20 ……… ………….

Declaration by Supervisor
This project has been submitted for examination with my approval as the supervisor

SIGNATURE…………………………… DATE……………………………

DR. JOSEPHINE NGURE

PURE AND APPLIED SCIENCES DEPARTMENT
SCHOOL OF PURE AND APPLIED SCIENCES
KIRINYAGA UNIVERSITY

i
Acknowledgement
We would like to acknowledge the almighty God for giving us the strength to carry out the study.
We thank our supervisor Dr. Josephine Ngure for her support and direction towards successful
completion of the project. Our family members have been encouraging to all of us and
supportive financially.

ii
Dedication
We would like to dedicate this project to our parents for their love, unwavering support and countless
sacrifices that they made to make this journey successful. Their encouragement has been our
motivation.

iii
Abstract
Setting of premiums is a common practice in many insurance companies that necessitates
utilization of the best actuarial models. However, many losses have occurred in some insurance
companies leading to collapse of the firms due to poor utilization of premium pricing models.
Boda boda insurance has been faced with the problem of ineffective pricing of premiums further
impacting its uptake in the country. This study used credibility models to effectively set
premiums for boda bodas. This study proposes that effective use of credibility models can
enhance setting of optimal premiums for both the insurer and the insured. Various credibility
models to be utilized in this study are the Bayesian credibility, Empirical Bayesian Credibility
Theory model 1 and model 2. The expected outcomes of this project is to establish credibility
models as the best in setting of effective premiums.

iv
Table of Contents
Declaration...................................................................................................................................................i
Acknowledgement.......................................................................................................................................ii
Dedication...................................................................................................................................................iii
Abstract......................................................................................................................................................iv
CHAPTER ONE: INTRODUCTION...................................................................................................................2
1.1 BACKGROUND INFORMATION...........................................................................................................2
1.2 Statement of the Problem.................................................................................................................4
1.1 Objectives of the Study................................................................................................................4
Main objective.....................................................................................................................................4
specific objectives................................................................................................................................4
1.4 Limitations of the Study.....................................................................................................................5
1.5 Scope of the Study.............................................................................................................................5
CHAPTER TWO: LITERATURE REVIEW..........................................................................................................6
2.1 Introduction.......................................................................................................................................6
2.2 Empirical Bayesian credibility theory.............................................................................................7
2.3 Empirical Review................................................................................................................................7
2.4 Conceptual Framework................................................................................................................8
CHAPTER THREE: METHODOLOGY...............................................................................................................9
3.0 Introduction.................................................................................................................................9
3.1.3 Methods of data analysis.......................................................................................................11
3.2 Credibility theory.............................................................................................................................11
3.2.1 Basic credibility model..............................................................................................................12
3.2.2 Bayesian Credibility Approach..................................................................................................13
3.3.1 EBCT Model 1............................................................................................................................16
Parameter estimation........................................................................................................................17
3.3.2 EBCT Model 2............................................................................................................................18
Specifications.....................................................................................................................................18
The Credibility Premium........................................................................................................................19
Parameter estimation........................................................................................................................20
CHAPTER 4: DATA ANALYSIS AND RESULTS...............................................................................................23
4.1 Credibility Premium.........................................................................................................................23
4.1.1 Results of model 1....................................................................................................................23

v
CHAPTER 5: CONCLUSION AND RECOMMENDATIONS..............................................................................27
5.1 Conclusion.......................................................................................................................................27
5.2 Recommendation............................................................................................................................27
REFERENCES..............................................................................................................................................28

vi
CHAPTER ONE: INTRODUCTION
1.1 BACKGROUND INFORMATION
Insurance can be defined as a contract or a policy in which an insurer protects a policy holder
against losses from specific perils. It is where an insured agrees to pay some premiums to be
covered by an insurance company in case any underlying risk occurs in a specified period of
time. Insurance covers aims at minimizing or eliminating certain measurable risk of economic
loss.
Insurance business in Kenya traces all the way back to the colonial period when Kenya was a
British state. The most common insurance companies include; Jubilee Insurance Company, Pan
Africa Insurance, Pioneer Assurance Society, and Provincial Insurance Company Limited.
Presently, the insurance business in Kenya is regulated by a body known as Insurance
Regulatory Authority (IRA) and member companies are regulated by the Association of Kenyan
Insurers (AKI) (Origin of protection business in Kenya, 2012).
There are many assets in Kenya that require insurance due to possible risk of loss. Some of these
assets include motor vehicles, motor cycles (Boda Bodas), and buildings. There is increased
number of motor cycles in Kenya especially the densely populated towns. Due to increased use
of boda bodas in the country, there is increased number of accidents necessitating insurance
covers for the losses that occur. However, few boda boda riders have been interested in taking
insurance covers compared with other types of covers like fire and motor vehicle. The insurance
companies tend to charge higher premiums due to increased number of risks in motor cycle
industry.

Various factors are considered by insurance companies before agreeing to cover motor cycles
(boda bodas). Some of the factors that may be considered include location, age and riding
experience, frequency of use, type of the bike, and accident and claim history. Regardless of
these factors, the amount of premiums set has not been favorable for more boda bodas to take.
There exists little or no research because there are few motor cycle riders taking insurance
covers. Most researches have mainly established there is fewer number of motor cycle riders
taking insurance covers. In their research, (karau, 2015) found that among the 172 motor cycle
crashes assessed in some rural parts of Kenya, only a third of them were insured. While there are
many risks factors promoting motor cycle accidents there is need to ensure a higher number of

7
individuals take insurance covers. A claim in motor insurance refers to the demand of payment
of a loss (in form of property damage or loss of life) by a policyholder or by an injured third
party. In reference to The Insurance Act (2015), all claims must be settled within 90 days from
the claim report date. Faster claim settlement is becoming one of the top objectives of all
insurance companies because that would make them unique as it enhances their credit rating.

The insured, while reporting a claim, must be able to provide: Police report form, Pictures of the
accident, Driving License. Estimates of items damaged in the vehicle: Proof of Doctors’ report if
any. In the recent past, the payment of claims by some automobile insurance companies has
become a great challenge. Many insurance companies have been forced to go under statutory
management over the inability to settle outstanding claims owed to policyholders. Blue Shield
insurance was put under statutory management in 2011. This has left behind many claimants
without compensation while other policyholders were left uninsured and had to look for
alternative insurance policies. Companies that have been declared insolvent due to
unprecedented losses include Stallion insurance in 2010, Access insurance company in 2018, and
Concord Insurance Company in 2013.

This project sought to use credibility models to set effective premiums. This will prevent the
insurance firms from making underwriting losses on boda boda policy while giving favorable
premiums for boda bodas. It will enhance confidence among the boda bodas hence encouraging
more individuals to take more insurance covers. It will boost insurance uptake in the country.

8
1.2 Statement of the Problem
The risk modeling techniques used by insurance companies to determine the value of the
premium often tend to analyze one fundamental question “how likely is it that this specific
customer will make a claim?” A high-risk policy will have a higher premium compared to a low-
risk policy. Risk in this case refers to the exposure of a motorist to any probable loss involving
the insured motor cycles. Motor cycles are generally considered high risk policies.
Most boda bodas are reluctant to take insurance covers due to high premiums that have been
brought by increased number of risks in the boda boda industry. According to (karau, 2015),
motor cycles have become the most preferred means of transport in many developing countries
due their increased mobility in urban roads and other maneuvers. Despite the increased number
of motor cycles, the number of boda boda taking insurance covers is not growing. This study
aims to solve the problem of instability in the pricing boda insurance products by using the most
appropriate credibility models in the calculation of insurance premiums.
1.1 Objectives of the Study
Main objective
The main objective of this project was to determine the optimal premium for boda boda
insurance using Empirical Bayes credibility Approach.
specific objectives
The specific objectives for this project were

1. To determine the credibility factor using Bayesian credibility approach

2. To identify the most appropriate credibility model for calculating premiums for boda
boda insurance policy.
3. to propose a model to be used in the industry to calculate premiums for boda boda policy.

1.3 Significance of the Study

The significance of this study was to increase the uptake of insurance policies by boda boda
through offering optimal insurance premiums. It suggested the best credibility models to be used
in the calculation of optimal insurance premiums.

9
1.4 Limitations of the Study
The study had several limitations; Credibility models uses past data to price the most effective
prices for premiums. Acquiring the past data was a challenge as some organizations do not easily
share their data, The use of past data to make future predictions has not always been reliable
because factors affecting premiums may vary with time, Actuarial data useful for the calculation
of premiums has not been available for all regions within the county which may hinder effective
calculation of the premiums.
1.5 Scope of the Study
The use of Boda Bodas in Kenya has highly been increasing. There is influx of motor cycles in
big towns and cities in the country. The study targeted the boda boda within Nairobi region, one
of the leading in terms of boda boda use.

10
 CHAPTER TWO: LITERATURE REVIEW
2.1 Introduction
The cost of motor insurance premiums is a crucial factor for policyholders and insurers alike.
This review of the literature will examine the state of risk identification in the Kenyan bodaboda
insurance today, with an emphasis on the variables that affect premium calculation. The study
intends to enlighten insurers about the need for a more thorough and precise premium calculation
procedure by shedding light on current processes and pointing out potential gaps in risk
assessment.

Effective premium pricing is an issue of key concern in many insurance firms. Various models
have been utilized in the calculation of premiums. Some of the models have been effective in
obtaining optimal premium values for both the insured and the insurers. Some of the common
models that have been utilized in premium pricing include credibility theory, empirical Bayesian
Credibility Theory (EBCT). The chapter reviews literature on work before by other scholars on
pricing of boda boda insurance policy premiums. According to (Yan, 2017) credibility theory
provides a mathematical rigorous procedure for deciding how much credibility should be given
to the actual experience rating of an individual risk relative to the manual rating common to a
particular class of risks. · (Schmidli, 2018) · Preview · provides an overview of classical
actuarial techniques, including material that is not readily accessible elsewhere such as the
Ammeter risk model and the Markov-modulated risk model. (Yanyan , 2013) · Credibility theory
provides important guidelines for insurers in the practice of experience rating. (Lowry, Wilson,
& Haig, 2014) Websites are often the first or only interaction a consumer has with a firm in
modern commerce
(Hans , 2015) shows us that to meet the Basel II regulatory requirements for the Advanced
Measurement Approaches in operational risk, the bank's internal model should make use of the
internal data, relevant external data, scenario analysis and factors reflecting the mark
(Promislow, 2015) Preview · Provides an introduction to the mathematics of financial markets,
exploring options, risk-neutral evaluation, the fundamental theorem of asset pricing and the
Black-Scholes formula. Provides coverage of profit testing.

11
2.2 Empirical Bayesian credibility theory
This theory was formed by Robbins as a result of combination of Bayesian and credibility
models. This model has the limitation of theory and practice. It has the challenge of demanding
the knowledge of prior distribution which is a rare necessity in practice. This led to the
emergence of Normal-Normal and the Poisson/Gamma models. The Bayesian Credibility
theories did not have effective ways of determining input distributions. The sophisticated
computing power through the estimation of input distribution based on empirical data and
calculated using the computers.
2.3 Empirical Review
Currently, it has become necessary to do pricing of insurance products attractively in order to
gain a competitive edge. This impacts insurers’ pricing strategies, profit realizations, and ability
to compete. By embracing technological advancement, understanding how the market is and
knowing consumer’s complexities, insurers must adapt to better pricing techniques.
(Laryeal, 2016) used the empirical Bayesian credibility model to price premiums for motor
insurance in Ghana. He argued that large variability exists within the non-life insurance firms
claim data implying that there is a very subtle sector responsive to variations.
The use of EBCT provided a consistent estimate of premiums signifying the use of the actuarial
tools is the only way to go. In their research, (Jindrova & Kopecka, 2017) applied Buhlmann-
Straub model to compute credible net premiums for insured catastrophic events in seven regions.
To cover the insured catastrophic losses, the estimates of credible net premiums by regions were
needed. This indicated how much money was needed to cover the events of losses. They found
that in the event of risk, enormous loss will occur. Therefore, there was a need to price the
policies such that when the risk occurs, the company can cover the losses.
(Singei, 2015) used Buhlmann and Buhlmann-Straub in determining the premiums for
health insurance claims. However, he argued that the Buhlmann and the Buhlmann-Straub
models met the hitch of outliers that might alter the mean and variance data affecting the
accuracy of the credibility premiums. This depicts the need for a better model to increase the
accuracy of the credibility premiums calculated.

12
 2.4 Conceptual Framework
Determining the proper cost of insurance coverage for motor vehicles requires a sophisticated
review of multiple elements, which is involved in premium pricing. Insurance companies
evaluate a variety of factors to determine the degree of risk attached to each policyholder and
motor vehicle. There are many variables that are considered when pricing premiums. The leading
determinants are: Type of the motor cycle; The type of motor cycle determines the price. Motor
bikes from different companies have different materials and engines that determine the insured
declared interest. For instance, motor cycles with diesel engines have higher premiums compared
to those with petrol engines, Location; Location is main factor considered when determining
premiums. Some areas are highly prone to risks of loss necessitating increased premium rates.

For instance, some areas have higher level motor cycle thefts necessitating higher premium
values compared to areas with lower levels of risk, Driver-related Factors: Age and Gender:
Young and inexperienced drivers may pose higher risks. Driving History: A clean driving record
usually results in lower premiums. Annual Mileage: Higher mileage increases the likelihood of
accidents and claims. Statisticians have revealed that young men are more prone to accidents
compared to young women. Young men are, therefore, given higher premiums than young
women, Frequency of use; Motor cycles with higher frequency of use have a probability of
making accidents. Individuals with higher frequencies of motor cycles use are given higher rates
of premiums while those who ride occasionally receive lower premiums.

Factors relating to vehicles: Make and Model: Certain cars are more likely to be involved in
collisions or theft. Year of Manufacture: Repair expenses for newer cars might be greater. Safety
Features: Discounts may be available for cars with cutting-edge safety features. Type of
Coverage: Type and Scope of Coverage: Rates are affected by the amount of comprehensive,
collision, and liability coverage, Allowable deductions: Lower rates are often the consequence of
higher deductibles. Information for Policyholders: Credit History: In some areas, premiums are
set based on credit scores, Occupation: Some occupations could have fewer risks than others.

13
Driving Habits and Behavior: Telematics Data: Insurers may use data from telematics devices to
monitor driving habits, Usage-based Insurance: Policies that consider actual vehicle usage
patterns. History of Claims: Past Claims; Higher premiums may be the consequence of a history
of frequent claims. Discounts and Incentives: Multi-Policy Discounts: Combining auto and home
insurance can result in discounts, Good Student Discounts: Offered to students with high
academic achievements. Regulatory and Legal Requirements: Local Regulations: Compliance
with regulatory requirements influences premium pricing. Market Conditions: Competitive
Landscape: The pricing strategies of competitors in the market, Overall Economic Conditions:
Economic factors can impact premium levels. Technological Advancements: Advancements in
Risk Assessment: The use of AI, machine learning, and data analytics for more accurate risk
evaluation. Actuarial Analysis: Statistical Models: Actuaries use statistical models to assess risk
and set premium rates.

CHAPTER THREE: METHODOLOGY

3.0 Introduction
This chapter specifies the models used in this study giving a description of the best models used
in the calculation of optimal insurance premiums. This chapter also shows the research design,

3.1 Research Design

We intent to adopt descriptive research design in this research.

3.1.2 Methods of data collection

3.1.2.1 Data on past claims

14
Collecting historical claims data for motor cycles is crucial. This includes information on
accidents, damage, and the associated costs. This data provides insights into the frequency and
severity of past events, forming the basis for risk assessment in this study.

3.1.2.2 Insurance Claims Records:

Accessing existing insurance claims records allows for the extraction of valuable data on past
bodaboda insurance claims. This includes information on the type of claims, the rates of the
amount paid, and the circumstances surrounding each claim.

3.1.2.3 Reports from the government and traffic safety:

Data regarding accident rates, road conditions and regulatory changes can be obtained by looking
through official reports and databases on traffic safety. The inclusion of more extensive trends
and patterns in the model from an external data source will improve its forecasting ability.

3.1.2.4 Customer demographic information:

For us to enhance the complexity of risk profiles, demographic data about motorbike owners,
including age, gender, location, and driving experience, can be gathered. External sources or
customer records may provide this information too.

3.1.2.5 Vehicle Specifications and Characteristics:

Collecting detailed information about the motor cycles being insured, including make, model,
year, and safety features, will be essential in this study. This helps us in assessing the inherent
risk associated with different types of motor cycles.

3.1.2.6 Social Media Observation:

Further qualitative data can be obtained by searching social media channels for conversations on
motorbike accidents, insurance claims, and safety issues. This approach will provide us an
instantaneous view of public opinion.

3.1.2.8 Financial and Economic Indicators:

15
Incorporating economic indicators, such as inflation rates, interest rates, and GDP growth, will
be relevant when assessing the broader economic context that may impact insurance rates.

3.1.3 Methods of data analysis

Bayesian Credibility Analysis:

We applied Bayesian statistical methods to estimate the credibility weights for various risk
factors. Bayesian analysis incorporates prior knowledge and updates it based on observed data,
allowing for a more dynamic and flexible modeling approach.

Calibration of Models:

We compared the expected and actual results of the Bayesian model to calibrate it. In order to
increase the model's accuracy in representing the observed data, this requires adjusting the
model's parameters.

Estimating Parameters:

We estimated the parameters of the Bayesian model, such as variance and mean and use
statistical techniques. The accuracy of parameter estimate is essential to the predictability of the
model.

Comparing Models:

To determine which Bayesian model or variant is best, we compared several of them. To

improve prediction accuracy, this entailed comparing models with various prior distributions or
adding more variables.

3.2 Credibility theory

Credibility theory is a statistical technique developed by Thomas Bayes, used to calculate pure
premium for the future based on the past data. The premiums are calculated as a weighted
average estimate of the past claims experience.

16
Methods for Determining the Optimal Premium

Basic Credibility Model; Basic credibility models estimate premiums based on individual claim
experience, taking into account the risk profile of each policyholder.

Bayesian Credibility Model; Bayesian models combine individual claim data with population-
based information, improving accuracy and reducing the impact of limited claim experience.
Simulation-based Credibility Model; Simulation-based models utilize advanced statistical
simulation techniques to estimate premiums, providing insurers with robust pricing strategies.

3.2.1 Basic credibility model

Basic credibility model is a statistical technique developed by Thomas Bayes, used to calculate
pure premium for the future based on the past data. The premiums are calculated as a weighted
average estimate of the past claims experience. Z is a quantity whose value is between 0 and 1,
which is taken as the weight, and used to determine the amount of impact that an estimate has on
the premiums calculated.

The credibility premium formula is given by:

Z X +(1−Z) μ, 0 ≤ Z ≤1…………………(1)

Where Z is the credibility factor which determines the level of trust attached in the data from the
actual risk.

X Indicates the expected aggregate claims/number of cases for the following year dependent on
data from the actual risk.

μ Indicates the expected aggregate claims/number of cases for the following year dependent on
collateral data, for example data from related risks not really indistinguishable from, the specific
risks viable.

Value of Z depends how reliable the estimates from data are. The higher value of Z the more the
impact the X estimate has on the calculated premium while the less of an impact the μ estimate
and vice versa.

Reliability of data is affected by how similar the risks and period of time over which the data
spans.

17
Our problem now is how to calculate Z. There are two methods to these problems: Bayesian
credibility and empirical Bayes credibility theory.
3.2.2 Bayesian Credibility Approach
It involves estimating Z by assuming a particular probability distribution for the claim
experience and estimating a parameter θ that is also a random variable.

It involves the following steps:

It begins with a prior distribution for the unknown parameter under deliberation (i.e., the claim
occurrence). It summarizes any information we have about its probable values. The prior
distribution should, therefore, be derived based on the collateral data provided, the next step is to
collect the appropriate data to derive the likelihood function, then find the posterior distribution
by combining the likelihood function and the prior distribution; A loss function is indicated to
convey how severe miscalculating the parameter value would result in.
The loss function should be constructed on viable reflections of the financial impact if the
insurance organization incorrectly estimates the parameter (and therefore the premium amounts).
Then Z is calculated as the value of Bayesian estimate of the parameter.
We shall demonstrate this approach using two different models: The Poisson/Gamma model and
the Normal-Normal model.

3.2.2.1 Poisson/ Gamma model

The Poisson/Gamma model is used to predict claim frequency by supposing that the number of
claims has a Poisson distribution with parameter λ. The value of λ follows a Gamma (α , β )
distribution. The random variable X signifies the number of claims in the following year from the
risk. The distribution of X depends on the fixed, but unknown, value of a parameter λ . The
conditional distribution of X given λ is Poisson ( λ ) while the prior distribution of λ is Gamma (α ,
β ).
The prior parameter distribution of λ is;

α
β α−1 −βλ
f prior (λ)= λ e λ¿ 0 …………….(2)
Г (α )

X 1 , X 2 , … , X n are preceding identified values of X. (It will be denoted by X for convenience

purposes)

18
The likelihood function that can be gotten from these data values is:

e−λ λ x e−λ λ ∑ x
n i i

L=∏ =
i=1 xi ! ∏ xi !
The posterior distribution is obtained by:

f post ( λ) ∝ f prior (λ) . Likelihood

e−λn λ∑ x β α α −1 −βλ
i

f post ( λ)= ∙ λ e …………………….(3)

∏ x i Г (α )
The posterior distribution is Gamma (∑ x i +α, n+β) while the Bayesian estimate of the unknown
parameter λ is the mean of the posterior distribution,

^λ= ∑ x+ α
n+ β

From credibility formula (1)

Z
∑ x +(1−Z) α
n β

n
Where Z=
n+ β

α
Z indicates the weighted average of the prior mean and the maximum likelihood estimate
β
∑x .
n

3.2.2.2 The normal/normal model

The normal model has been used to estimate the pure premium for specified risks. Letting X to
be a random variable representing the aggregate claims in the coming year for this risk. We can
make the following assumptions: The distribution of X depends on the fixed, but unknown, value
of a parameter, θ, the conditional distribution of X given µ is N (µ, σ 21) where σ is known. The
prior distribution of µ is N (θ, σ 22).

Therefore, PDF of the prior distribution is given by:

19
 ………(4)

The values of µ, σ 1 and σ 2 are known.

n past values of X have been observed, which will be denoted X 1 , X 2 , … , X n, or, more briefly, X.
The likelihood function gotten from the data is:

…….(5)

Posterior distribution is given by:

f post (Ɵ) ∝ f prior ¿. Likelihood

The posterior distribution of θ given X is:

( )
2 2 2 2
µ σ 1+ n σ 2 x σ1 σ2
N ,
σ 21 +n σ 22 σ 21+ n σ 22

Where:

n
xi
X =∑
i=1 n

And:

2 2
µ σ 1+ n σ 2 x
E(θ| x ¿ = 2 2
σ 1 +n σ 2

20
 2 2
σ1 n σ2
= 2 2
µ+ 2 2
x
σ 1 +n σ 2 σ 1 +n σ 2

= Z x +(1−Z )µ

n
2
Where: Z = σ1
n+ 2
σ2

3.3 Empirical Bayesian Credibility Theory

Unlike Bayesian credibility EBCT makes no assumptions about the distribution of claim sizes.
Instead of estimating a parameter θ, EBCT estimates a function of θ, m(θ). EBCT doesn’t
assume a statistical distribution for θ.

There are two models of EBCT, model 1 and model 2.

3.3.1 EBCT Model 1

It is a generalization of the normal/normal model. It is used to estimate pure premiums or
perhaps claim occurrence for a risk given past data for n years.

Let Xj:(X1, …,Xn), denote the aggregate claims or the number of claims in successive periods for
a risk. This value will be denoted by X . The values Xj, are identically distributed.

The following are the assumptions that ought to be made for EBCT Model 1

1. The distribution of each X j depends on a parameter, denoted θ, whose value is fixed (and the
same for all the X j s) but is unknown.
2. Given θ, the X j’s are independent and identically distributed.

For the risk parameter, θ, we define:

m(θ)=E(Xj|θ)

s2(θ)=var(Xj|θ)

EBCT model 1 results as follows:

(1 - Z)E[m(θ)] + Z X

21
Where:

n
Xj
X =∑
j=1 n

E[m(θ )] = μ

And:

n
Z=
n+ E [ s ( θ ) ] /var [m ( θ ) ]
2

Parameter estimation
In this part we are tried to estimate m(θ ) given X which will be completed by showing how to
estimate E[m(θ )] , var[m(θ )] and E[s2(θ )] .
We were interested in estimating the pure premium or the expected number of claims for a
particular risk, and that this risk is one of N risks. Let Xij represent the aggregate claims or the
number of claims for risk number I, i=1,2,…,N, for the year j, j=1,2,…,n.

The values can be summarized as;

Year

1 2 3 … n

1 X11 X12 X13 … X1n

Risk

Number 2 X21 X22 X23 … X2n

… … … … … …

N XN1 XN2 XN3 … XNn

The risk numbers could refer to different companies, insurers, shops, etc.

We define;

22
 n
X i =∑ X ij /n
j=1

And;

N
Xi
X =∑
i=1 N

The credibility premium for each risk is given by:

Z X i + ( 1−Z ) E [ m ( θ ) ]

Where:

n
Z=
n+ E [ s ( θ ) ] /var [m ( θ ) ]
2

The best linear parameter estimates are:

N
Xi
E [ m (θ ) ] =∑
i=1 N

( )
N n
1 1
E [ s ( θ ) ]= ∑
2
∑ (X ij− X i)2
N i=1 n−1 j=1

( )
N N n
1 1 1
var [ m ( θ ) ] = ∑
N −1 i=1
( X i−X )2− ∑
Nn i=1 n−1
∑ (X ij− X i)2
j=1

3.3.2 EBCT Model 2

EBCT model 2 is a generalization of model 1. The difference between EBCT model 1 and EBCT
model 2 is that EBCT model 2 takes into account the risk volume denoted by Pj .

Specifications
We need to estimate the expected aggregate claims in the following year for the specified risk.
Letting Y1, Y2,…, Yn be amount of claims in n successive years for a risk, our mission will be to
estimate Yn+1.
Introducing a new parameter known as risk size, Pj

23
The value of Pn+1 during the beginning of the year n + 1 is presumed to be known.
Next, a fresh classification of random variables, X1, X2… is defined as follows:
X j = Yj / Pj j = 1, 2,…
The random variable X j represents the aggregate claims, or the number of claims, in
Year j standardized to remove the effect of different levels of business in different years.

Assumptions of model 2

1. The distribution of each Xj depends on the value of a parameter, , whose

value is the same for each j but is unknown.

2. Taking, the Xj ’s to be independent (but not essentially identically distributed).

3. E(Xj | θ ) is not depended on j.

4. Pj var(Xj | θ ) is not depended on j .

Define:

m(θ)=E ( W j∨θ )

2
s (θ)=P j var ( W j∨θ )

such that:

E ( Y j )=P j m(θ)

2
Var ( Y j ) =P j s (θ)

E ( W j )=m(θ)

2
P j var ( W j ) =s (θ)

The Credibility Premium

We need to estimate the mean (given θ) of Yn+1, i.e. Pn+1 m(θ ).
The finest linear estimator of m(θ ) given X is:

24
 E [ m ( θ ) ] E[ s ( θ ) ]
2

n
var [ m ( θ ) ] + ∑ Y j
j=1
m ( θ )= n 2
E [s ( θ ) ]
∑ P j+ var [ m ( θ ) ]
j=1

It follows that;

Z X + (1- Z)E[m(θ )]

where:

n n

∑ Pj X j ∑Y j
j =1
X= n
= j=1
n

∑ Pj ∑ Pj
j=1 j=1

And

∑ Pj
j=1
Z= n
E [s 2 ( θ ) ]
∑ Pj+ var [ m ( θ ) ]
j =1

The outcome above indicates how to estimate the value of Xn+1 .In order to estimate Yn+1 , the
aggregate claim amount or claim occurrence for the following year, we need to multiply our
estimate of Xn+1 by Pn+1 , the risk size for preceding year.
Parameter estimation
The process for approximating the parameters E[m(θ )] , var[m(θ )] and E[s2(θ )] for EBCT
Model 2 is similar that of EBCT model 1.

N is the number of sampled risks; i.e. Accident, Theft, Fire and Damage from riots.

Year

25
 1 2 3 … N

1 Y11,P11 Y12,P12 Y13,P13 … Y1n,P1n

Risk

Number 2 Y21,P21 Y22,P22 Y23,P23 … Y2n,P2n

… … … … … …

N YN1,PN1 YN2,PN2 YN3,PN3 … YNn,PNn

Let Yij be a random variable indicating the aggregate amount claims for Risk Number i in Year j
, j = 1, 2,…, n , i = 1, 2,…, N , and let Pij be the resultant risk volume.
For each i and j define:
Xij = Yij / Pij

Define:

n
Pi=∑ Pij
j=1

N
P=∑ P ij
i=1

( )
N
1 Pi
P=
¿
∑
Nn−1 i=1
Pi 1−
P

n
Pij X ij
X i =∑
j=1 Pi

N
P i X i N n Pij X ij
X =∑ =∑ ∑
i=1 P i=1 j=1 P

The unbiased estimators are given as:

26
 E [ m (θ ) ] =¿ X

( )
N n
1 1
E [ s ( θ ) ]= ∑
2
∑ P ij ( X ij −X i )2
N i=1 n−1 j=1

( )
N n
1 1
var [ m ( θ ) ] = ¿ ∑ ∑ P ij ( X ij −X )2 −E [ s 2 ( θ ) ]
P Nn−1 i=1 j=1

27
 CHAPTER 4: DATA ANALYSIS AND RESULTS

Trends in claims
100000
99000
98000
Amount in Kshs

97000
96000
95000
94000
93000
92000
1 2 3 4 5
Years

Figure 1: Aggregate claims for Boda Bodas from 2019 to 2023

Key for figure 1
Year 1- 2019
Year 2- 2020
Year 3- 2021
Year 4- 2022
Year 5- 2023

The graphs indicate an inconsistent pattern in the number of claims. However, in 2023 the
number of claims seems to be higher compared to other years. It may indicate a growing
economy with a greater number of individuals venturing into the boda boda industry.
4.1 Credibility Premium
We discuss the results obtained from EBCT model 1 and model 2.
4.1.1 Results of model 1
n n 2
Policy w ij 1 ( wi−w )
∑ n
=wi ∑
n−1 j=1
( w ij−wi )
2

j=1

Boda boda 5592.6 1110266 1232690590756

Commercial 5828.6 763546.3 574136112893

28
 PSV 7839.8 1798531 3206573899343
Comprehensive MV 5037.6 670960.3 443453042375
Third Party MV 4267.2 415502.7 169114636460
Total 22973.2 3648540 5625968281827

Table 1.1: Summary of mean and variance of the policies

The overall mean of the policies is:
4
wi
E [ m (θ ) ] =∑
i=1 N

22973.2
E [m ( θ ) ]=
5
E [ m (θ ) ] =μ=¿5743.3

The average of variance of each policy is:

( )
N n
1 1
E [ s2 ( θ ) ]= ∑ ∑ (w −w i)2
N i=1 n−1 j=1 ij

( )
4 5
1 1
E [ s ( θ ) ]= ∑ ∑ (wij −wi)2
2
4 i=1 4 j=1

E [ s2 ( θ ) ]=¿ 912135
The overall variance is:

(∑ )
N N n
1 1 1
var [ m ( θ ) ] = ∑
N −1 i=1
( wi−w )2− ∑
Nn i=1 n−1
(w ij−wi )2
j=1

(∑ )
4 4 5
1 1 1
var [ m ( θ ) ] = ∑
3 i=1
( w i−w )2− ∑
20 i=1 4
(wij −wi)2
j=1

var [ m ( θ ) ] =¿2177397

We now calculate the credibility factor :

n
Z=
n+ E [ s ( θ ) ] /var [m ( θ ) ]
2

29
 5
Z=
912135
5+
2177397
Z=¿ 0.9226947
The pure premium for the year 2024 for boda boda from the credibility formula is therefore;
P=Z W +(1−Z )μ
P = 0.9226947 (5592.6) + (1-0.9226947) * 5743.3
P = 5604.25
4.1.2 Results for model 2

( )
Policy Pi Pi
Pi 1−
P
Commercial 1468 1114
PSV 1483 1122
Comprehensive MV 1561 1161
Third Party MV 1573 1166
Total 6085 4562
P*= 285
Table 1.2: Total volume for each policy
Next we calculate w i∧w
Policy wi
∑ Pij ( w ij−wi )
2
∑ Pij ( x ij−w ) 2
Commercial 5123 19952788 671139448.6
PSV 7310 107778217 9438537612
Comprehensive MV 4801 710636 501816055.2
Third Party MV 3866 24098821 3641898602
E((m ( θ )) = 5234

And,

( )
N n
1 1
E [ s ( θ ) ]= ∑
2
∑ P ij ( wij −wi)2
N i=1 n−1 j=1

30
 ( )
4 5
1 1
E [ s2 ( θ ) ]= ∑ ∑ Pij (wij −wi )2
4 i=1 4 j=1

E [ s2 ( θ ) ]=249286965
Also,

( )
N n
1 1
var [ m ( θ ) ] = ¿ ∑ ∑ P ij (wij −w i) −E [ s ( θ ) ]
2 2

P Nn−1 i=1 j=1

( )
4 6
1 1
var [ m ( θ ) ] = ∑ ∑ P (w −wi )2 −E [ s 2 ( θ ) ]
131.71146 23 i=1 j=1 ij ij

var [ m ( θ ) ] =¿ 1702264
The credibility factor for Boda Boda is given by:
n

∑ Pj
j=1
Z= n
E [s2 ( θ ) ]
∑ P j + var m (θ )
j=1 [ ]
6085
Z=
249286965
6085+
1702264
Z = 0.9947352
The pure premium per unit volume of Boda Boda for the year 2024 according to EBCT model 2
is calculated as follows:
P=Z w+(1−Z) μ
P=0.896018(5592.6)+ ( 1−0.896018 ) 5603
P=¿5593.595

31
 CHAPTER 5: CONCLUSION AND RECOMMENDATIONS
5.1 Conclusion
The main objective for this project was to identify effective credibility formula when calculating
premiums for boda bodas. This objective has been achieved identifying EBCT II as the best
model. The premiums from each model as the following:
Model

Premiums
EBCT I 5604

EBCT II 5594

Obtaining appropriate premiums for all insurance users should be the aim of every insurance
company. The models have obtained almost similar values. The model to be utilized when
calculating premiums may depend on the cost of implementation.

5.2 Recommendation
This project research proposes EBCT II as the most appropriate. EBCT II utilizes business
volume (weights) which enhances accuracy when calculating the most appropriate premiums.
This study proposes that effective use of EBCT II can enhance setting of optimal premiums for
both the insurer and the insured.

32
REFERENCES
Hans , B. (2015). Advanced Measurement Approaches in Operational risk.

Jindrova, & Kopecka. (2017). Buhlmann-Straub Model.

karau. (2015).

Laryeal, P. (2016).

Lowry, P. B., Wilson, D., & Haig, B. (2014). Websites for consumer interaction in Modern Commerce.

Promislow, D. S. (2015). Introduction to Mathematics of Financial Markets.

Schmidli, H. (2018). Overview of classical actuarial techniques.

Singei. (2015).

Yan, Y. Y. (2017). Mathematical rigorous procedure for decidng credibility.

Yanyan , z. (2013). Important guidelines for insurers in the practise of experience rating.

Yuje , Y. Y. (n.d.).

33
5.3APPENDICES
5.3.1 Analysis

Trends in claims
100000
99000
98000
Amount in Kshs

97000
96000
95000
94000
93000
92000
1 2 3 4 5
Years

Figure 1: Aggregate claims for Boda Bodas from 2019 to 2023

Results of model 1

n n 2
Policy w 1 ( wi−w )
∑ nij =wi ∑
n−1 j=1
( w ij−wi )
2

j=1

Boda boda 5592.6 1110266 1232690590756

Commercial 5828.6 763546.3 574136112893
PSV 7839.8 1798531 3206573899343
Comprehensive MV 5037.6 670960.3 443453042375
Third Party MV 4267.2 415502.7 169114636460
Total 22973.2 3648540 5625968281827

Results for model 2

( )
Policy Pi Pi
Pi 1−
P

34
 Commercial 1468 1114
PSV 1483 1122
Comprehensive MV 1561 1161
Third Party MV 1573 1166
Total 6085 4562
P*= 285

The value of w i∧w

Policy wi ∑ Pij ( w ij−wi )
2
∑ Pij ( x ij−w ) 2
Commercial 5123 19952788 671139448.6
PSV 7310 107778217 9438537612
Comprehensive MV 4801 710636 501816055.2
Third Party MV 3866 24098821 3641898602
E((m ( θ )) = 5234

Calculation of Premiums

Model

Premiums
EBCT I 5604

EBCT II 5594

35