BiUNIVERSITY OF AGRICULTURE, FAISALABAD

Department of Mathematics and Statistics

Faculty of Sciences
(Synopsis for M Phil degree in Mathematics)

TITLE: Analysis the comparison of Bayesian and Maximum Likelihood

Estimation Techniques for a Shape Parameter in an
Informative Prior Distribution.

Name of the Student : …….. …

Registration No. : .. … … …

Abstract
Statistical inferences are made with the Bayesian method, technique that uses the
Bayes' theorem to revise the likelihood of a given hypothesis when there is a new influx of
data, Bayesian inference is a fundamental concept in statistics, particularly mathematical
statistics method. An estimator for decision-making that minimizes the posterior anticipated
value of a loss function is known as a Bayes estimator in estimator theory and decision
theory. It optimizes the utility function's posterior expectation. The investigations will
explore the implications of various informative priors on the robustness and efficiency of
Bayesian estimators to derive corresponding posterior distributions. The work will delve
deeper into the exploration of alternative loss functions, aiming to comprehensively assess
their impact on Bayes estimators and associated risks. This prospective study anticipates
providing valuable insights into the effectiveness of different priors and loss functions in
Bayesian analysis, contributing to the refinement and enhancement of statistical inference
methodologies.
 UNIVERSITY OF AGRICULTURE, FAISALABAD
Department of Mathematics and Statistics
Faculty of Sciences
(Synopsis for M Phil degree in Mathematics)

I. TITLE: Analysis of Bayesian Estimation Techniques for

Informative Prior Distribution by Using Different Loss
Functions
II. a) Date of admission : 15-10-2022

b) Date of Initiation : 31-8-2023

c) Probable Duration : 4 Months

III. PERSONEL:
a) Name : … … … ..
b) Regd. No. : …………..

IV. SUPERVISORY COMMITTEE:

V. Introduction
The Binomial and Hypergeometric distributions are the other two discrete
distributions that use integers as random variables; the Informative Prior distribution is one of
them. The Informative Prior distribution was found by Simeon-Denis Informative Prior in
1838 as an approximation to the binomial distribution. When there is a high number of trials
and a low chance of success, it is usually utilized. Though they have many chances to
happen, informative prior occurrences are very uncommon. For this reason, the "law of small
numbers" is frequently used to the Informative Prior distribution.

The Bayesian statistical method, on the other hand, makes use of Bayes' theorem to
update a statistical model's parameters in response to observed data. The posterior
distribution is obtained by combining the observational data in the form of probability
functions with the prior distribution, which symbolizes the previous knowledge. Thereafter,
future events may be predicted using the posterior distribution. The stages involved in
Bayesian analysis will be covered in this article, from building the prior and data models to
drawing conclusions, verifying the model, and improving it. We will also go into the
significance of variational inference, sampling from posterior distributions, prior and
posterior predictive analysis, and variable selection techniques (van de Schoot et al., 2021).

As a simpler estimate method, Han introduced E-Bayesian estimation. Jahren and

Okasha studied the E-Bayesian estimate for the Burr Type XII model, which focuses on type-
II censoring. He considered applications with error prediction capabilities, such as Bayes, E-
Bayes, and strong Bayes predictions for future discovery, as a preventive strategy.

Gonzalez-Lopez offered tenable solutions that centered on E-Bayesian

approximations and exponential distribution for device performance. Yousef Zadeh
introduced the stability of the apparatus to determine parameters for E-Bayesian and
hierarchical Bayesian estimations based on the function with the asymmetric loss. To get
estimators of E-Bayesian for structural approach-based statistical data, Okasha and Wang
computed the E-Bayesian estimates for a variety of distributions. (Mohammed, n.d.).

That loss function should be used for option value model estimation and analysis?
There have been a number of different goals put forth, but no clear winner has emerged as of
yet. We emphasize that choosing loss functions must be continuous. First, for each model, the
loss function used to measure the parameters and assess the model should be the same in
order to prevent getting less-than-ideal parameter estimations. Secondly, in order to prevent
drawing inappropriate correlations, the estimate loss function for each model should be the
same (Christoffersen and Jacobs, 2004).

One of the discrete distributions that are used the most is the Informative Prior
distribution, which is used since data from many different research disciplines frequently
conform to its postulates. Using only one parameter as a basis, however, restricts its
adaptability in a variety of implementations. Data over dispersion concerning the Informative
Prior distribution is a common problem (Breslow, 1990; Dean, 1992). Allowing it is assumed
that the Informative Prior parameter is a random variable with a probability distribution,
which results in a hierarchy of distributions, is one way to get around this restriction
(Macedo, 1948; Satterthwaite, 1942).By using an Informative Prior parameter gamma mixing
distribution in a common hierarchy, the observed data have a negative binomial distribution.
Data are under-dispersed compared to the Informative Prior distribution in several solutions,
such as word texts' and dictionaries' lengths. In other words, there are various uses with
empirical distributions with broader or shorter tails compared to the Informative Prior
distribution, where ratio sequential probabilities changes with time (Shmueli et al., 2005).

Although the multidimensional version of the Informative Prior distribution has been
utilized significantly less frequently in practice due to computational difficulties with
inferential approaches, it has been routinely used to explain solitary count data. When there
are a lot of zero counts and little counts, it might be difficult to apply an approximation
multivariate Informative Prior model utilizing a multivariate normal model. The multivariate
Informative Prior distribution is the most significant discrete multivariate distribution;
however, it has a number of limitations linked to applications. The main shortcoming of the
multivariate Informative Prior model is its complicated joint probability function
representation. As a consequence, pairwise comparisons of variables have been performed
using a simplified model that has a single common covariance component for all variables.
(Karlis and Meligkotsidou, 2007).

Objectives

The objectives of this study are:

1. To derive corresponding posterior distributions, future investigations could explore
the implications of various informative priors on the robustness and efficiency of
Bayesian estimators.
2. To delve deeper into the exploration of alternative loss functions.
3. To comprehensively assess their impact on Bayes estimators and associated risks.
Review of Literature
Yu and Meyer (2006) suggested a publicly accessible BE named Win BUGS can
readily do a completely likelihood-based estimate. Nine multivariate Stochastic Volatility
(SV) models with Granger causality in volatility are included. The specifications that permit
variable correlation coefficients over time appear to be best, according to empirical evidence.

Das and Roy (2011) proposed weighted generalized Rayleigh distribution (WGRD).
Length biased has been used to generate familiar distribution. The length-biased of WGRD
has been observed to be an appropriate distribution for modeling monthly mean minimum
temperature pattern data after actual application to authentic data.
 Ahmed et al. (2013) presented size biased Gamma distribution (SBGD). A SBGD has
been defined, which is a type of weighted Gamma distribution (WGD) that uses weights as
variable values. A HE has additionally been gotten.

Aslam et al. (2014) suggested Fréchet distribution (FD) using Bayesian paradigm.
The Posterior distribution (PD) has been solved using quadrature numerical integration
approach. Prior predictive distributions have calculated. A MC simulation study has used to
compare the performance of BE.

Reshi et al. (2014) introduced sized biased distribution of generalized Gamma

distribution (SBDGGD). This distribution structural features have discussed in detail. Using
BE for a unique model, parameter estimation has achieved.

Xu et al. (2014) presented BE objective priors have been used to estimate the number
of species in a population. Two objective prioresses on the mixed-Informative Prior. We
conclude that reference prior is better than the Jeffreys' rule prior.

Reyad et al. (2016) suggested QE-Bayesian estimation which has a variation of E-

Bayesian estimation approach. QE-Bayesian estimation to original E-Bayesian method in
predicting FD scale parameter. The QE-Bayesian estimation and E-Bayesian estimation
characteristics have been also investigated.

Saraiva and Suzuki (2017) presented Metropolis-Hastings and Slice sampling

methods for predicting WD parameters. The Metropolis-Hastings algorithm was created with
two versions independent Metropolis-Hastings and random walk Metropolis. The sample root
means square errors and bias have been used to compare techniques.

Sultana et al. (2017) presented three-component mixture of FD. BE of the parameter

of the mixture model, as well as their posterior risks, have been generated using various non-
informative and informative priors. The statistical features of BE have been simulated for
varied sample sizes.

Abbas et al. (2019) suggested medical data has frequently been filed for each
patient to help in decision-making. Skewed distributions might be good candidates for
deriving conclusions using BE. The current work focuses on the building of BE for three-
parameter FD utilizing noninformative prior and gamma prior.
 Yousaf et al. (2019) proposed transmuted distributions have been a skewed family of
flexible distributions. To estimate the size and shape parameters of the aforementioned
distribution. A simulation study has been undertaken using the Markov Chain MC
technique for uncensored and censored settings.

Bodnar et al. (2020) worked on BE to estimate three deter-mining parameters of the

efficient frontier. Posterior distribution has been computed by applying diffuse and conjugate
priors. BE for all three parameters has supplied together with standard uncertainties and
asymptotic distributions.

Arias et al. (2021) proposed a broad framework for BE and causality assessment.
Priors and samples from PD using a MC. MC technique to assess the probability of an all-
purpose epidemiological model are the cornerstone of our approach. In terms of production,
we have decreased mortality at no cost or a very low cost.

Long (2022) proposed number of units seeing event has random but the study overall
time is fixed a type-I filtering mechanism emerges. A variety of mathematical algorithms
have been created. The frequentist maximum likelihood and BE estimate three parameters of
FD.

Material and Method

Different loss functions will be used to find BE of Informative Prior Distribution

Gaussian Distribution

The Gaussian Distribution is defined as

e−( x− μ) /2 σ
F(x| μ ¿ = , t=0 , 1 ,2 , … , μ>0 (1)
√2 π σ 2
Gamma Distribution

The prior distribution with gamma conjugate for μ, as

θ
λ − λ μ θ −1
π ( μ )= e μ ; λ¿ 0 , θ>0 , μ>0 (2)
Γθ

Stein Loss Function (SLF)

Estimated value ^μ of parameter μ, then SLF is defined as

 μ^
L ( μ , ^μ ) = −ln
μ
μ^
μ
−1() (3)

Power Loss Function (PLF)

Estimated value of ^μ of parameter μ, then PLF is defined as

μ^^ μ
L ( μ , ^μ ) = + −2 (4)
μ μ^^

K- Loss Function (KLF)

Estimated value of of parameter , then KLF is defined as

^μ μ

√ √
2
μ ^
μ
L ( μ , ^μ ) =( − ) (5)
μ^ μ

DeGroot Loss Function

Estimated value of ^μ of parameter μ, then DeGroot loss function is defined as

μ−^μ
L ( μ , ^μ ) = (6)
μ^

References

Abbas, K., B. Gilani, S. Sabir, M. Altaf and F. Shahzad. 2019. Analysis of Frechet
Distribution Based on Type-I Censored Data. Tech. J. 24:98–104.

Ahmed, A., J.A. Reshi and K.A. Mir. 2013. Structural properties of size biased Gamma
distribution. IOSR J. Math. 5:55–61.

Arias, J., J. Fernández-Villaverde, J. Rubio Ramírez and M. Shin. 2021. Bayesian Estimation
of Epidemiological Models: Methods, Causality, and Policy Trade-Offs.

Aslam, M., S.M.A. Kazmi, I. Ahmad and S.H. Shah. 2014. Bayesian estimation for
parameters of the Weibull distribution. Sci. Int. 26:1915–1920.

Bodnar, T., M. Lindholm, V. Niklasson and E. Thorsén. 2020. Bayesian Quantile-Based

Portfolio Selection. arXiv Prepr. arXiv20:2012-01819.
Christoffersen, P. and K. Jacobs. 2004. The importance of the loss function in option
valuation. J. financ. econ. 72:291–318.

Das, K.K. and T.D. Roy. 2011. Applicability of length biased weighted generalized Rayleigh
distribution. Adv. Appl. Sci. Res. 2:320–327.

Karlis, D. and L. Meligkotsidou. 2007. Finite mixtures of multivariate Informative Prior

distributions with application. J. Stat. Plan. Inference 137:1942–1960.

Long, B. 2022. Estimation and prediction for Topp–Leone distribution using double Type-I
hybrid censored data. J. Stat. Comput. Simul. 1–21.

Mohammed, H.S. n.d. Empirical E-Bayesian estimation for the parameter of Informative
Prior distribution.

Reshi, J.A., A. Ahmed, K.A. Mir and S.P. Ahmad. 2014. Bayesian analysis of size-biased
generalized gamma distribution–a simulation study. Int. J. Mod. Math. Sci. 10:45–59.

Reyad, H.M., A.M. Younis and S.O. Ahmed. 2016. QE-Bayesian and E-Bayesian estimation
of the Frechet model. Br. J. Math. Comput. Sci. 19:62–74.

Saraiva, E.F. and A.K. Suzuki. 2017. Bayesian computational methods for estimation of two-
parameters Weibull distribution in presence of right-censored data. Chil. J. Stat. 8:25–
43.

Shmueli, G., T.P. Minka, J.B. Kadane, S. Borle and P. Boatwright. 2005. A useful
distribution for fitting discrete data: revival of the Conway–Maxwell–Informative Prior
distribution. J. R. Stat. Soc. Ser. C (Applied Stat. 54:127–142.

Sultana, T., M. Aslam and J. Shabbir. 2017. Bayesian analysis of the mixture of frechet
distribution under different loss functions. Pakistan J. Stat. Oper. Res. 501–528.

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 Prim. 1:1–26.

Xu, C., D. Sun and C. He. 2014. Objective Bayesian analysis for a capture–recapture model.
Ann. Inst. Stat. Math. 66:245–278.

Yousaf, R., M. Aslam and S. Ali. 2019. Bayesian estimation of the transmuted Frechet
distribution. Iran. J. Sci. Technol. Trans. A Sci. 43:1629–1641.
Yu, J. and R. Meyer. 2006. Multivariate stochastic volatility models: Bayesian estimation and
model comparison. Econom. Rev. 25:361–384.