ON THE NORMAL MOMENT DISTRIBUTIONS

Similar documents
ON A GENERALIZATION OF THE GUMBEL DISTRIBUTION

Distributions of Functions of Random Variables. 5.1 Functions of One Random Variable

On q-gamma Distributions, Marshall-Olkin q-gamma Distributions and Minification Processes

Continuous Random Variables

Tail dependence in bivariate skew-normal and skew-t distributions

The Type I Generalized Half Logistic Distribution Downloaded from jirss.irstat.ir at 11: on Monday August 20th 2018

4. Distributions of Functions of Random Variables

Multivariate Distribution Models

Chapter 4 Multiple Random Variables

Multivariate Normal-Laplace Distribution and Processes

Random Variables and Their Distributions

Statistics for scientists and engineers

p-birnbaum SAUNDERS DISTRIBUTION: APPLICATIONS TO RELIABILITY AND ELECTRONIC BANKING HABITS

arxiv: v1 [math.pr] 10 Oct 2017

Chapter 5 continued. Chapter 5 sections

Modelling Dependence with Copulas and Applications to Risk Management. Filip Lindskog, RiskLab, ETH Zürich

BMIR Lecture Series on Probability and Statistics Fall 2015 Discrete RVs

Geometric Skew-Normal Distribution

Multivariate Distributions

Univariate Normal Probability Density Function

2 Functions of random variables

Continuous Distributions

Spherically Symmetric Logistic Distribution

Some results on Denault s capital allocation rule

Topic 4: Continuous random variables

Miscellaneous Errors in the Chapter 6 Solutions

Topic 4: Continuous random variables

3. Probability and Statistics

Joint p.d.f. and Independent Random Variables

Part IA Probability. Definitions. Based on lectures by R. Weber Notes taken by Dexter Chua. Lent 2015

Math 3215 Intro. Probability & Statistics Summer 14. Homework 5: Due 7/3/14

Generation from simple discrete distributions

Introduction to Statistical Inference Self-study

A NEW CLASS OF SKEW-NORMAL DISTRIBUTIONS

Statistics, Data Analysis, and Simulation SS 2015

Sampling Distributions

International Journal of Scientific & Engineering Research, Volume 6, Issue 2, February-2015 ISSN

1 Review of Probability and Distributions

MAS223 Statistical Inference and Modelling Exercises

Continuous Random Variables

Multivariate Statistics

Statistics 3657 : Moment Generating Functions

1.12 Multivariate Random Variables

Bivariate Sinh-Normal Distribution and A Related Model

1 Probability and Random Variables

Probability and Distributions

Stat 5101 Notes: Brand Name Distributions

Multiple Random Variables

Probability. Machine Learning and Pattern Recognition. Chris Williams. School of Informatics, University of Edinburgh. August 2014

Modulation of symmetric densities

MA/ST 810 Mathematical-Statistical Modeling and Analysis of Complex Systems

Review 1: STAT Mark Carpenter, Ph.D. Professor of Statistics Department of Mathematics and Statistics. August 25, 2015

Review: mostly probability and some statistics

Deccan Education Society s FERGUSSON COLLEGE, PUNE (AUTONOMOUS) SYLLABUS UNDER AUTOMONY. SECOND YEAR B.Sc. SEMESTER - III

PCMI Introduction to Random Matrix Theory Handout # REVIEW OF PROBABILITY THEORY. Chapter 1 - Events and Their Probabilities

Sampling Distributions

A THREE-PARAMETER WEIGHTED LINDLEY DISTRIBUTION AND ITS APPLICATIONS TO MODEL SURVIVAL TIME

Continuous random variables

On prediction and density estimation Peter McCullagh University of Chicago December 2004

Chapter 3 sections. SKIP: 3.10 Markov Chains. SKIP: pages Chapter 3 - continued

Cheng Soon Ong & Christian Walder. Canberra February June 2018

ECE 4400:693 - Information Theory

On Bayesian Inference with Conjugate Priors for Scale Mixtures of Normal Distributions

STAT 801: Mathematical Statistics. Distribution Theory

On Five Parameter Beta Lomax Distribution

Foundations of Statistical Inference

Contents 1. Contents

S6880 #7. Generate Non-uniform Random Number #1

Some generalisations of Birnbaum-Saunders and sinh-normal distributions

Let X be a continuous random variable, < X < f(x) is the so called probability density function (pdf) if

Lecture 17: The Exponential and Some Related Distributions

Stat 5101 Lecture Slides: Deck 8 Dirichlet Distribution. Charles J. Geyer School of Statistics University of Minnesota

3 Continuous Random Variables

Gaussian Processes. Le Song. Machine Learning II: Advanced Topics CSE 8803ML, Spring 2012

P (x). all other X j =x j. If X is a continuous random vector (see p.172), then the marginal distributions of X i are: f(x)dx 1 dx n

Review of Statistics I

Testing a Normal Covariance Matrix for Small Samples with Monotone Missing Data

Exam 2. Jeremy Morris. March 23, 2006

Introduction to Normal Distribution

MCMC 2: Lecture 3 SIR models - more topics. Phil O Neill Theo Kypraios School of Mathematical Sciences University of Nottingham

, find P(X = 2 or 3) et) 5. )px (1 p) n x x = 0, 1, 2,..., n. 0 elsewhere = 40

Stat 5101 Notes: Brand Name Distributions

Chapter 5. Chapter 5 sections

Probability Distributions

From Determinism to Stochasticity

Lecture 3. Probability - Part 2. Luigi Freda. ALCOR Lab DIAG University of Rome La Sapienza. October 19, 2016

2. The CDF Technique. 1. Introduction. f X ( ).

EEL 5544 Noise in Linear Systems Lecture 30. X (s) = E [ e sx] f X (x)e sx dx. Moments can be found from the Laplace transform as

x. Figure 1: Examples of univariate Gaussian pdfs N (x; µ, σ 2 ).

04. Random Variables: Concepts

Continuous Random Variables. and Probability Distributions. Continuous Random Variables and Probability Distributions ( ) ( ) Chapter 4 4.

Probability Distribution And Density For Functional Random Variables

Bayes Estimation and Prediction of the Two-Parameter Gamma Distribution

APPM/MATH 4/5520 Solutions to Exam I Review Problems. f X 1,X 2. 2e x 1 x 2. = x 2

IE 230 Probability & Statistics in Engineering I. Closed book and notes. 60 minutes.

Weighted Exponential Distribution and Process

Test Problems for Probability Theory ,

Partial Solutions for h4/2014s: Sampling Distributions

Multivariate Random Variable

APPM/MATH 4/5520 Solutions to Problem Set Two. = 2 y = y 2. e 1 2 x2 1 = 1. (g 1

Transcription:

ON THE NORMAL MOMENT DISTRIBUTIONS Akin Olosunde Department of Statistics, P.M.B. 40, University of Agriculture, Abeokuta. 00, NIGERIA. Abstract Normal moment distribution is a particular case of well known Kotz-type elliptical distributions, the density which depends on a shape parameters α, such that α = 0 corresponds to the normal probability density function; several important properties of this class of density function and relationship with existing distribution are established. Keywords and phrases: Normal Moment Distribution, Cumulative Distribution Function,Probability Density Function, Generalized Birnbaum-Saunder Distribution, Reflected Gamma Distribution. AMS(000) Mathematics Subject Classification:Primary 60E05. Introduction Some univariate distributions that have been of great importance lately is the family of elliptical contour densities or simply elliptical distributions. This family includes distributions with lesser or greater kurtosis than the normal distribution. Furthermore, the elliptical family has the normal distributions as a special case. The elliptical laws have been studied by numerous authors and most important results were obtained by Fang,Kotz, and Ng [990]. The use of elliptical distribution as a generalization of normal distribution is not based on empirical arguments or on physical laws. In general, its reasoning is purely statistical and /or mathematical, in the sense that (a) the theory developed under normal distribution is a particular case of the theory derived within elliptical distributions; (b) many of the properties of the normal distribution can be generalized to the case of the elliptical distributions; (c) some important statistics in the theory of normal inference are invariant within the elliptical family. for these reasons, currently, a large part of normal theory is being reconstructed using elliptical distributions, allowing that any statistical analysis in which a normal distribution All correspondence to be addressed to this author e-mail:- akinolosunde@yahoo.ca

can be assumed, can be generalized to this whole family. For a random variable (one dimensional case), the elliptical distributions are symmetrical distributions in R. A r.v. X with elliptical distribution is characterized by the parameters µ and σ and is characterized by a function generator of densities,g, for which the notation X is EC(µ, σ;g) is used. In general, µ and σ are position and scale parameters respectively. µ=e(x) when the first moment of the distribution exists. Var(X)=c 0 σ when the first two moments exists, where c 0 = φ (0) and φ is the derivative of the function φ associated with characteristic function of X. in this article we introduce parameter α which is the shape parameter such that α= 0 corresponds to normal distribution and we called the distribution normal moment distribution. The normal moment distribution can be derived from Kotz-type distribution see Fang et al [990]. Let X be continuous r.v. such that X is (0,;g). Thus, the characteristics function of X is given by ψ(x) = φ(x ); x R, () with φ:r + R, and the p.d.f. of X is given by f(x) = cg(x ); x R, () where g is the kernel of the pdf of X and c the normalization constant. X follows the Kotz-type symmetric distribution with parameters r, s > 0 and q > / and the pdf is f(x) = xr(q )/s Γ((q )/s) x(q ) exp( rx s ), x R. (3) Where normal distribution is a particular case of the Kotz type when q=s= and r=/ in this work we found the pdf of normal moment by introducing a shape parameter α which is define below starting from Kotz-type distribution.. Definition A r.v. X from standardized pdf in (4)above is said to be normal moment distribution if r=/, s= and q=α+. Therefore, X is a normal moment r.v. if it has pdf given by f(x) = e x α+ Γ(α + < x <, α 0 (4) )xα Where α is the shape parameter and Γ(.) is gamma function. Then we say that X is a normal moment random variable with parameter α; for brevity we shall say that X is NM(α), (4) is a proper density function because;if f:r R > 0 denote a positive real valued-function. We say f is very rapidly decreasing (VRD) if every smooth compactly supported function g is a uniform limit (on R) of functions of the form f(x)p(x), where P(x) is a polynomial, thus we have this theorem; Theorem. If f(x) is VRD, then the moment integrals; f(x)xα dx converges for all α. 3

Proof. The space of smooth compactly supported functions is of infinite dimension, and all such spaces are uniform limits of sequences f(x)p α (x). Therefore, f(x)p(x) L (R) for some polynomial P of arbitrary large degree. Therefore, f(x)= (x α ) for all α, and the lemma follows.thus, x α e x < x <, α 0 (5) exists and the integrand is a positive continuous function which is bounded by an integrable function. Also, the normal moment distribution can be attributed as a special case of the Sobolev (q-r)family of life distribution which is widely used in fatigue life modeling Sobolev (994). The importance of the statistical theory and application of the normal distribution is well known. Since the proposed distribution is more flexible than the normal distribution, it is anticipated that this class of the distribution will give a better fit for some sets of data for a selected values of the shape parameter α, in this paper we shall consider only standardized type that is where X is NM(α,0,). The first section have introduced the topic and some literature review. In second section some discussions about the shape of the distribution are presented. The third section; the even moment, properties and the relationship with well known distributions are established.the last section shows the estimation of the parameters of the distribution and finally some remarks on this distribution. SHAPE One of the unique properties that distinguished normal moment distribution is the shape of the pdf, it can easily be seen that dlogf(x)/dx= α, see (figure.0. Appendix A).We notes that as α increases the degree of peakedness also increases, f(0)=0 and 0 f(x)= 0 f(x), also, the shape is similar to reflected gamma distribution introduced by Borghi(965) which includes Laplace distribution as a special case when the shape parameter α=. The cumulative distribution function is not in close form the tables of approximate values for some selected α has been produced by Olosunde (007) using numerical computation method. 3 THE MOMENTS AND RELATED DISTRIBU- TIONS In what follows we derived the rth moment of the distribution associated even rth moment is given as If X is NM(α) then the E(x r ) = xr f(x)dx = r Γ(α + r + ) Γ(α + ) (6) 4

it can thus be seen that E(X) = 0 (7) var(x) = α + (8) γ = 0 (9) γ = α + 3 (0) The expressions γ and γ are the skewness and kurtosis respectively for the standard normal moment distribution. We establish some properties and prove theorems that shows relationships between the normal moment distribution and some well known distributions. Theorem 3. Let X be a continuous random variable having the even normal moment distribution and α 0. Then the following properties hold; (a) NM(0)=N(0,), (b) if X is NM(α) then -X is NM(α), (c) X =Y is χ (α+). Proof: (a)and(b) are obvious, (c) can easily be proved that; P[Y y]= P[X y]=p[- y X y] F (y) = y 0 r/ Γ(r/) zr/ e z/ dz, for y > 0, () and when r=α+,the result follows Theorem 3. Let X and X be two independent continuously distributed random variable each having the even normal moment distribution symmetric about 0 with parameters α and α respectively, then the random variable F = X /(α + ) X/(α + 0 < f < () ), has an f-distribution with r = α + and r = α + degree of freedom. Proof: and h(x ) = k(α )x α e x / (3) h(x ) = k(α )x α e x / (4) 5

from equation () above since x is it can be shown that f( α + α + )p, x = p (5) h F,P (f, p) = J h X,X [g (f, p), g (f, p)], (6) where J is the jacobian of the transformation which is nonzero and from theorem above we have define a new random variable and propose finding the marginal p.d.f. g (f) of F,given the equation for the joint p.d.f. g(f,p)of the random variables f and x =p that is; g(f, p) = f α + / α +α Γ(α + )Γ(α + )(α α + / )α +/ p α +α + e p [+f( α +/ α +/ )], (7) then the marginal pdf g of F is f r / ( r r ) r / Γ( r +r ) Γ( r )Γ( r )[ + f( r r )] (r +r )/ 0 < f < (8) This is the pdf of f-distribution with degree of freedom r =α + and r =α +. Theorem 3.3 Let X be a random variable of continuous type having the normal moment distribution symmetric about 0 and α=. then Y = ln x has Gumbel pdf having the shape parameter λ= Proof: it can be shown that f Y (y) = d dy g (y) f X (g (y)) e y e e y, forλ =, < y <, (9) The corresponding moment generating function is t Γ( t) (0) This is the moment generating function of Gumbel distribution the shape parameter λ=. Theorem 3.4 Let X and X be two stochastically independent random variables which are N(0, ) and normal moment distribution respectively, then Y = x x has Cauchy-type distribution. 6

proof: Given the joint density function of X, X and y as stated above, for y = x, the transformation such that the joint pdf of y and y is, f Y,Y (y, y ) = J f X,X (g (y, y ), g (y, y )), < y <, < y < = f(y, y ) = k(α) π y α e y and marginal pdf of y say g (y ) and solving, we have (+y ) () k(α) π 0 y α+ e y (+y ) dy = Γ(α + ) Γ(α + ) π ( + y ) α+ () This is Cauchy type distribution, note that when α = 0 the distribution reduces to Cauchy distribution. and g (y ) does not converge for α=0, where the µ r moment is Γ(r + )Γ(α r + ) Γ(α + ) (3) 4 ESTIMATION We consider estimation by the method of maximum likelihood. from (4) introducing a new variable of integration, say y, by writing then (4) becomes f(y; α, µ, σ) = α+/ Γ(α + /) x = y µ, σ > 0, (4) σ (y µ) α σ α+ The log-likelihood for a random sample x,...x n n exp[ / ( y µ i= σ ) ], < µ <, σ > 0, < y <. (5) logl(α, µ, σ) = n(α+ n )ln nlnγ(α+ )+α ln(x i µ) n(α+)lnσ n (x i= σ µ). i= (6) The first order derivatives of (6) with respect to the three parameters are: lnl = n n(α+) i= + (xi µ), where σ σ σ 3 ni= (x ˆσ i µ) =, (7) n(α + ) 7

lnl α = n i= ln(x µ) nlnσ nln Γ (α+ ) Γ(α+ ) and lnl µ = σ ni= (x i µ) α n i= (x i µ). Setting the expressions to zero give the equations as stated above, the case of µ and α can be solve using numerical computation.γ (.) represent digamma function. 5 Remarks In this work, we have discussed a wider extension of the normal distribution starting with Kotz-type distribution, obtaining its density and some of its graphics to see how the shape parameter influences its behavior. We have also outlined some important properties of this new model, some of its anticipated usefulnes is in fatigue life modeling following the recent work of Diaz-Garcia and Leiva-Sanchez(005) where they presented the generalization of Birnbaum-Saunder distribution(gbs),which is denoted by T GBS(δ, β; g), then we have this theorem that: Theorem 5. Let T GBS(δ, β; g). Then, the pdf of T is given by f T (t, δ) = Γ(α + )( δ )α+ t ( β + β t )α exp[ δ ( t β + β )][t 3/ t (t + β) ]. (8) δβ / proof: This result is simply proof following the same procedure used in Diaz-Garcia and Leiva-Sanchez(005) References [] Allan, J.Mathematical Formulars and Integrals. Academic Press, Inc. [] Allan,S., and Keith, J. (985)Kendall s Advanced Theory of Statistics. Charles Griffin and Co. Ltd. London. [3] Balakrishnan,N., and Kocherlakota, S. (985). On the double weibull distribution: Order Statistics and estimation, Sankhya, Series B, 48, 439-444. [4] Borghi,O.(965).Sobre una distribucion de frecuencias, Trabajos de Estadistica,6, 7-9. [5] Diaz-Garcia, J.A. and Leiva-Sanchez,V.(005). A new family of life distributions based on the contoured elliptical distributions. Journal of Statistical Planning and Inference, 8(), 445-457. 8

[6] Fang,K.T., Kotz,S. and Ng, K.W.(990). Symmetric multivariate and related distribution. Chapman and Hall, London. [7] Olosunde, A.A., and Ojo, M.O. (007). Some properties of normal moment distribution, Ife Journal of Science, (To appear) [8] Sobolev, I.G.(994). Interrelations between probabilty distributions belonging to the q-r family, Izmeritel naya Tekhnika,0,-4. 9

Appendix A. Figure.0. The Pdf of Normal Moment Distributions for selected values of α=,,3 respectively and µ=0, σ =.