UNIVERSITY OF AGRICULTURE, FAISALABAD

Department of Mathematics and Statistics

(SYNOPSIS FOR M.Phil. IN MATHEMATICS)

TITLE: BAYESIAN APPROACH TO THE GENERALIZED EXPONENTIAL

DISTRIBUTION UNDER ASYMMETRIC LOSS FUNCTIONS
Name of the Student : Esha Tariq
Registration No. : 2021-ag-890

Abstract

The Generalized Exponential Distribution is a versatile and effective modeling tool in various
fields of study. This distribution can be used to model a wide range of phenomena, including
heavy-tailed and skewed data. Asymmetric loss functions are increasingly being used for
parameter estimation in this distribution, driven by its widespread adoption. In this study, we will
use the Bayesian approach to determine the unknown parameter of the Generalized Exponential
distribution under asymmetric loss functions such as the Stein loss function (SLF), Power loss
function (PLF), DeGroot loss function (DLF), Al-Bayyati loss function (ALF), and Minimum
Expected loss function (MELF) by taking gamma distribution into consideration as a conjugate
prior. A Monte Carlo (MC) simulation method will be used to identify the most suitable
asymmetric loss function for Bayesian inference of the Generalized Exponential Distribution.
The numerical results obtained from the Monte Carlo simulation by using R software will be
used to compare the efficacy of the estimation technique.
 UNIVERSITY OF AGRICULTURE, FAISALABAD
Department of Mathematics and Statistics
(SYNOPSIS FOR M.Phil. IN MATHEMATICS)
I. TITLE: BAYESIAN APPROACH TO THE GENERALIZED
EXPONENTIAL DISTRIBUTION UNDER ASYMMETRIC LOSS
FUNCTIONS
II a) Date of Admission 11-10-2021

b) Date of Initiation 14-09-2022

c) Probable Duration 04 Months

III PERSONNEL

a) Name of the Student Esha Tariq

b) Registration Number 2021-ag-890

IV SUPERVISORY COMMITTEE

i. Supervisor Dr. Muhammad Zafar Iqbal ([email protected])

ii. Member Mr. Khurrem Shehzad ([email protected])

iii. Member Dr. Muhammad Kashif ([email protected])

V INTRODUCTION
Bayesian inference is a strategy of statistical inference that updates the probability of a
hypothesis using Bayes' theorem whenever new evidence or information is made available. This
method was introduced by Reverend Thomas Bayes in 1763, In the field of statistics, particularly
mathematical statistics, Bayesian inference is an extremely important methodology. The
Bayesian analytical approach offers a logic-based context for quantitative and qualitative
components of data analysis (Reich and Ghosh, 2019).

The basic idea behind Bayesian analysis is simple and instinctive. We have several data
points to explain, along with a list of possible interpretations. The candidate explanations already
had some prior credibility as the best explanation before they were exposed to the new data.
When new information is supplied to us, we adjust our credentialing so that it places more
emphasis on candidate specifications that more accurately reflect the data and less emphasis on
candidate representations that do not accurately reflect the data. The rule of Bayes is a
mathematically convincing method for changing credentials. The rest are only insignificant
details (Kruschke and Liddell, 2018).

Bayesian statistics is a technique of data analysis in which the parameters of a statistical

model are upgraded with the help of observed data. Bayesian statistics was developed by George
Bayes. The past information known as prior distribution is paired with the observational data in
the form of a likelihood function to derive posterior distribution. Each parameter has a prior
distribution that captures uncertainty before observing the data. The likelihood function contains
all information and data about the model parameters Posterior can also be used to predict future
events (van de Schoot et al., 2021 ). This introduction will walk you through the steps of
Bayesian analysis, beginning with creating the prior and data models and continuing through
determining inference, model checking, and model refining (Van De Schoot and Depaoli, 2014).

The Bayesian analysis focuses on the fact that how the prior distribution affects the
posterior. Priors are determined by using techniques such as maximum entropy, reference priors,
and hierarchical priors. The results of inference are impacted by a weak or strong prior. Weak
priors may produce deceptive results, whereas strong priors may dominate likelihood and
produce biased results (Ghaderinezhad and Ley, 2019).

(Mahdavi and Kundu, 2017) introduce a method for increasing the flexibility of a
distributional group by including another parameter. A variant of the one-parameter exponential
distribution has been thoroughly examined. Logical expressions for the mode, moments, mean
residual lifetime, quantiles, moment-generating function, entropy, stochastic orders, and order
statistics are derived for the proposed distribution. MLE of unknown parameters cannot be
obtained explicitly; rather, they can be only obtained through the resolution of nonlinear
equations.

The exponential distribution was subjected to a generalized transformation, which

resulted in the GE distribution. The generalized exponential distribution is a flexible distribution
that may be employed to model several events in various disciplines, like hydrology, finance,
and reliability engineering. For assessing lifetime data, the generalized exponential distribution
(GE) may be substituted for the gamma distribution, Weibull distribution, and exponential
distribution. A Bayesian analysis is presented that uses Markov Chain Monte Carlo (MCMC)
calculations. In this analysis, different informative and non-informative prior distributions are
assumed to persist for the parameter of the model (Achcar et al., 2015).

Exponentiated exponential and exponentiated generalized exponential distributions are

sub-models of GE distribution. The GE distribution's statistical and mathematical features
present simple equations for the mean deviations, moments, and generating function. To estimate
the parameters of this distribution, the maximum likelihood technique is utilized.
This distribution is compared to some existing distributions for two real sets of data from the
fields of engineering and biology. This distribution is more appropriate and useful for studying
lifetime data and provides another alternative model for existing models (Ashraf et al., 2020).

The term ‘Loss Functions’ was introduced by Abraham Wald in 20th century (Wald,
1939). In Bayesian theory and statistical modeling, the difference between predicted and
observed values can be measured by using a mathematical function called loss function. The goal
of using a loss function is to reduce the error among actual and predicted values during training
so that the model can precisely predict outcomes on new, unseen data. By minimizing the loss
function, the model can learn to make more accurate predictions on new, previously unknown
information.

When the data have heavy-tailed distributions, the loss minimization and parameter
estimation in statistical learning investigate the behavior of various loss functions, including the
square loss and the absolute loss, and interpret their impact on parameter estimation (Hsu and
Sabato, 2016).

Objectives

The objectives of this study are:

1. To derive the posterior distribution by using informative prior.
2. To estimate the unknown parameters by using asymmetric loss functions.
3. To compare the efficiency of parameter estimation by using simulation methods.
VI REVIEW OF LITERATURE
Haddawy (1999) emphasized that Bayesian approaches are gaining popularity because of
their ability to handle uncertainty and provide flexible modeling frameworks. He highlighted the
importance of Bayesian problem-solving techniques and their capability to deal with some of the
most challenging situations in areas like healthcare, finance, and engineering. It also
emphasized the importance of further research and development to optimize the effectiveness
and applicability of these techniques in real-life conditions.

Elliott et al. (2003) addressed the significance of accurate parameter estimation in loss
functions, along with the limitations of existing methods for addressing the problem of
estimating loss function parameters. They then proposed a new method based on the
generalized method of moments framework for a more flexible and efficient estimation of loss
function parameters. They demonstrated the utility of their approach by presenting parameter
estimation examples for various loss functions using both simulated and real data.

Han (2017) demonstrated the relationship among E-Bayesian estimates by using three
different hyper-parameter prior distributions and presented an E-Bayesian estimation technique
for evaluating the probability of system failure.

Hossain (2018) proposed a new strategy by using MLE (maximum likelihood estimation)
method for estimating the parameters of the three-parameter generalized exponential distribution.
Afterward, the proposed method was compared to other methods in the literature, such as the
moment and probability-weighted moment methods. He used Monte Carlo simulations to
compare bias, MSE, and coverage probability to other methods. The simulation values showed
that the proposed MLE outperformed the other methods in most cases.

Dibal et al. (2019) studied the analysis of an unknown rate parameter of an Exponential
distribution that used a Bayesian framework under the Al-Bayyati loss function with non-
informative and informative prior distributions. They applied Bayes' theorem to build posterior
distributions for the unknown rate parameter and then used the Al-Bayatti loss function to
estimate the value of the parameter. The study observed that the estimates under various priors
were near the actual parameter, and the MSE increased as the rate parameter value increased for
all sample sizes. Additionally, the study found that the rate of Bayesian estimates by using
informative prior distributions were more efficient with a minimum MSE than non-informative
prior distributions.
 Kour et al. (2020) practiced Bayesian and E-Bayesian estimation techniques to estimate
the unknown scale parameter of the Exponential-Lomax Distribution by using conjugate gamma
prior distribution under the DeGroot, Minimum-Expected, and Al-Bayyati loss functions. They
used MATLAB to evaluate the computational data of the Bayesian and E-Bayesian estimates and
their associated mean squared errors (MSEs).

Mohammad et al. (2021) presented the generalized exponential Marshall-Olkin (GEMO)

distribution, which included a shape that was more adaptable than that of the GE distribution and
could more accurately quantify survival data. The probability density function (PDF), cumulative
distribution function (CDF), hazard function (HF), survival function (SF), moments, and mean of
the GEMO distribution were discussed, and the maximum likelihood technique was used to
evaluate parameter estimates.

Muhammad (2021) analyzed the shape parameter of the Lomax distribution by using
Bayesian estimation. The analysis was done with the use of loss functions under Jeffery and
uniform prior i.e., QELF, SELF, PELF, and the WSELF. Using simulated data under the
Bayesian framework, the effectiveness of multiple estimators is evaluated and represented
graphically. It was observed that the Bayes estimator with weighted error loss function (BWEL)
produces optimum results as compared to other methods under a uniform prior. The
precautionary error loss function (BPEL) yields adequate outcomes compared to other methods
under Jeffery's prior.

NJOMEN et al. (2021) presented the Bayesian estimation of competitive risks

and compared the efficacy of distinct loss functions in estimating model parameters. They did
simulation research to assess the effectiveness of relative loss functions in estimating model
parameters and then compared the results. They measure the effectiveness of the solution by
utilizing metrics such as mean absolute error, mean squared error, and mean LINEX loss. The
findings of the simulation study demonstrated that the selection of loss functions has the
potential to have a substantial influence on the accuracy of estimation.

Talhi and Aiachi (2021) conducted a Bayesian approach to the upper truncated
Zeghdoudi distribution based on type II censored data. They used different loss functions to
attain Bayes estimators and the posterior risks that corresponded to them. An approach that relies
on the Markov chain Monte Carlo (also known as MCMC) was proposed to investigate their
performance. Estimators with the maximum likelihood were obtained using the initial values
determined for the model's parameters. After contrasting the performance of these estimators to
that of Bayesian estimators using integrated mean square error and Pitman's closeness criterion,
the methodology was demonstrated using an example based on actual data.

Hussain et al. (2022) introduced a new distributions family known as the

generalized exponential extended exponentiated (NGE3) family and certain analytical
characteristics of the presented family were demonstrated. The maximum likelihood (ML)
method, also known as the method with the highest possible probability, was utilized for
parameter estimation. An uncommon model known as the NGE3-Exponential was investigated
and estimated with the help of ML. The performance of the estimators was demonstrated through
the usage of a simulation.

Nagy and Alrasheedi (2022) proposed that the Bayesian estimators and maximum
likelihood of the distribution's parameters, along with the hazard and reliability functions were
evaluated, by using Type I generalized progressive hybrid censoring data from the
generalized exponential distribution. These quantities' credible interval estimators were also
retrieved. Although some values could not be obtained in closed form, a Monte Carlo
computation technique with Gibbs sampling was adopted for simulation and analysis.

Kumar and Gupta (2023) utilized symmetric SELF and asymmetric loss functions by
using non-informative prior, to generate Bayes estimators for inverse Rayleigh distribution’s
unknown parameter. They adopted simulation to determine the comparative risk and efficiency
of the estimators and correlate their performance. According to the observations, the minimum
expected loss function outperformed the symmetric SELF, the precautionary loss function, and
the DeGroot loss function.

VII Materials and Methods

To obtain the Bayesian estimation of the Generalized Exponential distribution, various
types of loss functions can be utilized.

GENERALIZED EXPONENTIAL DISTRIBUTION

The generalized exponential distribution was invented by Gupta and Kundu (2007). It is a single
parameter ϑ distribution having probability density function as follows:

−x ϑ −1 −x
f ( x , ϑ )=ϑ (1−e ) e .

GAMMA DISTRIBUTION

Conjugate Gamma distribution as a prior distribution is given by,

a −1 −bϑ
π ( ϑ /a , b )=ϑ e .

POWER LOSS FUNCTION (PLF)

The estimated value of δ of parameter ϑ, then the power loss function is as followed:

δ ϑ
L ( ϑ , δ ) = + −2 .
ϑ δ

STEIN LOSS FUNCTION (SLF)

The estimated value of δ of parameter ϑ, then the Stein loss function is as followed:

δ
L ( ϑ , δ ) = + ln
ϑ
δ
ϑ ()
−1 .

DEGROOT LOSS FUNCTION (DLF)

The estimated value of δ of parameter ϑ, then the DeGroot loss function is as followed:

ϑ −δ
L (ϑ , δ)= .
δ

AL-BAYYATI LOSS FUNCTION (ABLF

The estimated value of δ of parameter ϑ, then the Al-Bayyati loss function is expressed as:

c1 2
L ( ϑ , δ ) =ϑ (ϑ −δ) .

MINIMUM EXPECTED LOSS FUNCTION (MELF)

The estimated value of δ of parameter ϑ, then the Minimum Expected loss function is expressed
as:

2
(ϑ−δ)
L (ϑ , δ)= 2
.
δ
VIII References

Achcar, J., F. Moala and J. Boleta. 2015. Generalized exponential distribution: A Bayesian
approach using MCMC methods. Int. J. Ind. Eng. Comput. 6:1-14.
Ashraf, J.H., M. Ahmad, A. Khalique and Z. Iqbal. 2020. A New Generalized Exponential
Distribution: Properties and Applications. Int. J. Anal. Appl. 18:799-818.
Dibal, N.P., A.T. Mobolaji and Y.A. Musa. 2019. Bayes’ estimators of an exponentially
distributed random variables using Al-Bayyati’s loss function. Int. J. Entific Res. Publ.
(IJSRP) 9:674-684.
Elliott, G., A. Timmermann and I. Komunjer. 2003. Estimating loss function parameters.
Soc. Sci. Res. Net. 404700.
Ghaderinezhad, F. and C. Ley. 2019. On the impact of the choice of the prior in Bayesian
statistics, Bayesian inference on complicated data. IntechOpen. p. 22.
Gupta, R.D. and D. Kundu. 2007. Generalized exponential distribution: Existing results and
some recent developments. J. Stat. Plan. Inference 137:3537-3547.
Haddawy, P. 1999. An overview of some recent developments in Bayesian problem-solving
techniques. AI Mag 20:11-11.
Han, M. 2017. The E-Bayesian and hierarchical Bayesian estimations for the system reliability
parameter. Commun. Stat. -Theory and Methods 46:1606-1620.
Hossain, S.A. 2018. Estimating the Parameters of a Generalized Exponential Distribution. J. Stat.
-Theory and Applications 17:537-553.
Hsu, D. and S. Sabato. 2016. Loss minimization and parameter estimation with heavy tails. J
Mach Learn Res 17:543-582.
Hussain, S., M. Sajid Rashid, M. Ul Hassan and R. Ahmed. 2022. The Generalized Exponential
Extended Exponentiated Family of Distributions: Theory, Properties, and Applications.
Mathematics 10:3419.
Kour, K., P. Kumar, P. Anand and J.K. Sudan. 2020. E-Bayesian Estimation of Exponential-
Lomax Distribution Under Asymmetric Loss Functions. int. j. appl. math. 59:14-39.
Kruschke, J.K. and T.M. Liddell. 2018. Bayesian data analysis for newcomers. Psychon. Bull.
Rev. 25:155-177.
Kumar, R. and R. Gupta. 2023. Bayesian Analysis of Inverse Rayleigh Distribution under Non-
Informative Prior for Different Loss Functions. Thail. Stat. 21:76-92.
Mahdavi, A. and D. Kundu. 2017. A new method for generating distributions with an application
to exponential distribution. Commun. Stat. - Theory Methods. 46:6543-6557.
Mohammad, R., D. Lestari and S. Devila. 2021. Generalized exponential Marshall-Olkin
distribution. In: J. Phys. Conf. Ser. p 012100.
Muhammad, I. 2021. Bayesian Estimation of the Shape Parameter of Lomax Distribution under
Uniform and Jeffery Prior with Engineering Applications. Gazi Univ. J. Sci. 34:562-577.
Nagy, M. and A.F. Alrasheedi. 2022. Estimations of generalized exponential distribution
parameters based on type I generalized progressive hybrid censored data. Comput Math
Methods Med 2022:1-15.
NJOMEN, D.A.N., T. DONFACK and D.W. TANGUEP. 2021. Bayesian estimation under
different loss functions in competitive risks. Glob. J. Pure Appl. Math. 17:113-139.
Reich, B.J. and S.K. Ghosh. 2019. Bayesian statistical methods. Chapman and Hall/CRC,
London.
Rossi, P.E. and G.M. Allenby. 2003. Bayesian statistics and marketing. Mark. Sci. 22:304-328.
Talhi, H. and H. Aiachi. 2021. On truncated Zeghdoudi distribution: posterior analysis under
different loss functions for type II censored Data. Pak. J. Stat. Oper.:497-508.
Van De Schoot, R. and S. Depaoli. 2014. Bayesian analyses: Where to start and what to report.
Eur. health psychol. 16:75-84.
van de Schoot, R., S. Depaoli, R. King, B. Kramer, K. Märtens, M.G. Tadesse, M. Vannucci, A.
Gelman, D. Veen and J. Willemsen. 2021. Bayesian statistics and modelling. Nat Rev
Methods Primers 1:1.
Wald, A. 1939. Contributions to the theory of statistical estimation and testing hypotheses. Ann.
Math. Stat 10:299-326.