![MathJax Logo](/templates/jsp/_style2/_tandf/pb2/images/math-jax.gif)
Abstract
In this paper, we consider a class of doubly truncated distributions. Based on this class, recurrence relation for moment generating functions of record values is derived. Recurrence relations for single and product moments of record values are deduced from the main results. Several of the known distributions can be seen as special cases of the used class, which includes the left, right and non-truncated distributions. So, we discuss in detail some examples of these distributions based on the main results. In general, for any distribution in the considered class, we reach to a general form to find recurrence relations of moment generating function and moments.
1. Introduction
There is a large volume of works based on the study of recurrence relations between moment generating functions of record values and characterizations based on these relations. The moments have considerable importance in the statistical literature. Many authors have studied and derived several identities and recurrence relations for the single and product moments. For example, from Burr Type XII distribution, Ahmad [Citation1] derived recurrence relations for single and product moments of record values and their characterizations. Ahmad and Fawzy [Citation2] studied the recurrence relations for moments of generalized order statistics (gos). Recurrence relation for moments of record values is derived by Kamps [Citation3]. Recurrence relations for moment generating functions of record values and order statistics are considered by AL-Hussaini et al. [Citation4]. For more details, the reader is referred to Saran and Singh [Citation5], Khan and Zia [Citation6], Kumar [Citation7] and Bashir and Ahmad [Citation8]. Record values have great importance since they use in several real-life problems involving economics, weather, industry and sports. It can be demonstrated as order statistics from a sample where its size is determined by the values and order of occurrence of the observations. Numerous papers have appeared discussing characterizations based on record values, such as, Balakrishnan and Balasubramanian [Citation9], Franco and Ruiz [Citation10,Citation11], Wu [Citation12,Citation13], Bairamov [Citation14]. For more details for survey of record values, see Ahsanullah et al. [Citation15] and Hassan et al. [Citation16]. In this paper, our focus is on recurrence relations for moment generating functions of record values from a class of continuous distributions in the doubly truncated case. Section 2 gives a recurrence relation for moment generating functions of record values, which can be applied to obtain relations for single and product moments of record values from this class. In Section 3, we apply such recurrence relation for moment generating functions of record values to characterize the doubly truncated class of continuous distributions.
Suppose that a random variable (rv) X has an absolutely continuous distribution function (df), as considered by AL-Hussaini [Citation17] and it is given by
so that the probability density function (pdf) is written as
where
is a nonnegative, strictly increasing and differentiable function of x such that
as
and
as
. The function
is the derivative of
with respect to x and represents the failure rate whereas
represents the corresponding survival function. We shall write the class of distributions as
(1)
(1)
A doubly truncated pdf on , denoted by
, is given by
(2)
(2) where
(3)
(3)
The doubly truncated df and survival function (sf) are given, respectively, for , by
(4)
(4) and
(5)
(5) where
,
is given by (4) and
(6)
(6)
Note that and
.
The doubly truncated class denoted by . Then, for
(7)
(7)
The non-truncated, right- and left-truncated classes are denoted by ,
and
, respectively. The right truncated class
is given by
(8)
(8) where
as
.
The left-truncated class is given by
(9)
(9) where
as
.
2. Recurrence relation for moment generating functions
Let , be k upper record values based on a sequence
of identically and independently distributed (i.i.d) rvs having pdf (2). The joint pdf of
,
, for
is given by
(10)
(10) where
(11)
(11)
For more details, see Arnold et al. [Citation18]. Note that (10) is based on k upper records, whereas the corresponding expression in the previous reference is based on k + 1 upper records (including the zeroth record). We shall use and
, to denote the joint moment generating function and joint moments of
, respectively, where
are nonnegative real numbers. For simplicity, we usually drop
in the computations and be satisfied with i, so, for example we write
instead of
. In other words,
(12)
(12) and
(13)
(13) where
is the joint pdf given by (10).
It may be observed that , then we can write a joint moment generating function as;
In the following theorem, a recurrence relation for moment generating functions from the doubly truncated df (4), is obtained for record values.
Theorem 2.1
Let be k upper records based on a sequence
of i.i.d rvs having the doubly truncated pdf (2). For nonnegative integers
the following recurrence relation for the moment generating function of
, is satisfied.
On the other hand, if condition (14) is satisfied, then the survival function corresponding to is given by (5). All of the expectations and moment generating functions involved are assumed to be exist.
Proof:
By replacing f and in (10) and (11) by
and
, respectively, the resulting joint
is used in definition (12) to obtain the joint moment generating function
By substituting in (16), we obtain
(17)
(17)
The first term in (17) is then , and Equation (17) becomes
From (2), (5) and (6), we have
Using this relation in the second term, we obtain
(18)
(18)
On the other hand, if condition (14) is satisfied, then using (18), (12) and (10), such condition may be rewritten as
or, equivalently,
Applying the extension of Müntz-Szász theorem [see, Hwang and Lin [Citation19]], we obtain
Thus has the pdf as given in (2).
Special cases
(1) By differentiating (14) with respect to and putting
, we obtain the following recurrence relation for product moments of record values
(19)
(19) where
(20)
(20) and
(2) If we put ,
, in (20) and, in addition,
, in (15), we obtain recurrence relations for the moment generating functions and moments for the single record value
. The moment generating function is given by
(21)
(21) where
(22)
(22)
This leads to the moment generating function of the record as
(23)
(23) where
and
is given by (22) after replacing
by j.
A recurrence relation for the moments of single upper records is given by
(24)
(24) where
(25)
(25)
This leads to the expectation of the , which can be written as follows
(26)
(26) where
and
is given by (25) after replacing
by j.
(3) If we put ,
, and
, in (14) and (20), we obtain recurrence relations for the moment generating functions and moments of
and
. The moment generating function is given by
(27)
(27) where
(28)
(28)
This leads to
(29)
(29) where
and
is given by (28) after replacing
by j.
A recurrence relation for the mixed moments is given by
(30)
(30) where
(31)
(31)
This leads to
(32)
(32) where
and
is given by (31) after replacing
by j.
Remark 2.1:
In all special cases of the doubly truncated distributions, recurrence relations remain the same, yet the following should be observed when computing the expectations.
In the non-truncated or left-truncated cases, where , (or
), we use
such that
for the non-truncated case and
for the left-truncated case.
In the right truncated case, we use the following equation in the computation of the expectations,
3. Examples
In this section, we shall apply previous results to write recurrence relations (RR) based on some members of class (4).
3.1. Doubly truncated Weibull distribution
Let and
.
Then the CDF of doubly truncated Weibull distribution is given by
(i) For which satisfy the conditions of the theorem, it follows from (15), so that we can write
From (20), the rhs of RR (19) is given, by
(ii) For it follows from (22) and it is easy to get that
From (25), the rhs of RR (24) is given by
(iii) For and using (28), we obtain
Also, using (31), we get
Remark 3.1:
(i) In the non-truncated case, . So that, the terms
reduce to 1. In this case, it follows from (22), so that we can write
It can be shown that
and
(33)
(33)
This is true since in the non-truncated Weibull case, we obtain
(34)
(34) where
is given by Arnold et al. [Citation19] as
In the non-truncated Weibull case, we can see that
By substituting in (34), we obtain
where
By substituting the values of and
in (23), we obtain a RR for the mgf of powers of single upper record values.
In the non-truncated Weibull case, the same RR (26) is obtained except that in this case,
It then follows, from (26), so we can write that
In this case, it can be shown that
This RR coincides with the RR obtained by Balakrishnan and Chan [Citation20] where they used
(ii) The exponential distribution is obtained by setting in the Weibull
distribution. That is, in the exponential
distribution,
.
(iii) The Rayleigh distribution is obtained by setting in the Weibull
distribution. That is, in the Rayleigh
distribution,
.
3.2. Doubly truncated compound Weibull (three-parameter Burr Type XII) distribution
Let
Then the CDF of doubly truncated compound Weibull distribution is given by
(i) For it follows from (15), we can obtain that
From (20), we have
(ii) For it follows from (22) and we can write
From (25), we obtain
(iii) For it follows from (28) and it is easy to see that
From (31), we have
Recurrence relations of the doubly truncated (Lomax and Rayleigh) distributions can be obtained by setting and
respectively, in the previous RR’s.
For example, to check the results, when ,
, we have the compound exponential (Lomax
) distribution. From RR (26), it is not difficult to get that
where
, for
. The non-truncated Lomax
distribution yields
Therefore, we get
This recurrence relation coincides with the recurrence relation obtained, differently, by Balakrishnan and Ahsanullah [Citation21] in which they used instead of
and
, respectively.
For the non-truncated Lomax distribution, it can be shown that
3.3. Doubly truncated Pareto (Type I) distribution (belongs to class =![](//:0)
)
Let
Then the CDF of doubly truncated Pareto distribution is given by
(i) For it follows from (15), so we can write that
From (20), we have
(ii) For and using (22), the rhs of RR (21) is given by
From (25), we obtain
(iii) For and using (28), we get
From (31), we have
It can be shown that RR (26) reduces to the following RR in the case of non-truncated Pareto I where
:
Therefore, we get
3.4. Doubly truncated beta distribution (belongs to class =![](//:0)
)
Let
Then the CDF of doubly truncated beta distribution is given by
(i) For it follows from (15), we get
We note that in the non-truncated case, since , then
.
From (20), we have
(ii) For and using (22), we get that
From (25), we have
(iii) For it follows from (28) and it is easy to get that
From (31), we have
In the non-truncated beta case (note that, in this case,
), RR (26) reduces to the RR
and it can be shown that
3.5. Doubly truncated Gompertz distribution
Let and
.
Then the CDF of doubly truncated Gompertz distribution is given by
(i) For it follows from (15), so we can get that
From (20), we have
(ii) For and using (22), we obtain that
From (25), we obtain
(iii) For it follows from (28), so we can get that
Using (31), we have
3.6. Doubly truncated compound Gompertz distribution
and
.
Then the CDF of doubly truncated Gompertz distribution is given by
(i) For and from (15), the rhs of RR (14) is given by
Using (20), the rhs of RR (19) is given by
(ii) Using (22) and for the rhs of RR (21) is given by
From (25), the rhs of RR (24) is obtained by
(iii) For and from (28), the rhs of RR (27) is given by
Using (31), the rhs of RR (30) can be written as
4. Conclusion
Based on a general class of doubly truncated distributions, recurrence relations for the moment generating functions and product moments of powers of upper record values are considered for a sequence of iid random variables. So, a general form in (14) can be used to find recurrence relations of moment generating function and moments of powers for any distribution includes in the class in (7) for any sequence of iid random variables. The considered class includes the right, left and non-truncated distributions as special cases. In addition, it includes the most important distributions than can be used in the life testing and other applied area of statistics. Some special cases are presented using (14) such as recurrence relations for moment generating function and moments of powers for univariate and bivariate upper record values. Also, the results of some of these special cases have been checked to ensure the agreement with already existing results in the literature.
Disclosure statement
No potential conflict of interest was reported by the author(s).
References
- Ahmad AA. Recurrence relations for single and product moments of record values from Burr Type XII distribution and a characterization. J Appl Stat Sci. 1998;7:7–15.
- Ahmad AA, Fawzy MA. Recurrence relations for single moments of generalized order statistics from doubly truncated distributions. J Stat Plan Inf. 2003;117:241–249. doi: 10.1016/S0378-3758(02)00385-3
- Kamps U. Recurrence relation for moments of record values. J Stat Plan Inf. 2002;45:225–234. doi: 10.1016/0378-3758(94)00073-5
- AL-Hussaini EK, Ahmad AA, EL-Boghdady HH. Recurrence relations for conditional moment generating functions of order statistics and record values. Discuss Math Prob Stat. 2005;25:181–205.
- Saran J, Singh SK. Recurrence relations for single and product moments of k-th record values from linear exponential distribution and a characterization. Asian J Math Stat. 2008;1:159–164. doi: 10.3923/ajms.2008.159.164
- Khan RU, Zia B. Recurrence relations for single and product moments of record values from Gompertz distribution and a characterization. World Appl Sci J. 2009;7:1331–1334.
- Kumar D. Recurrence relations for marginal and joint moment generating functions of generalized logistic distribution based on lower record values and its characterization. Prob Stat Forum. 2012;5:47–53.
- Bashir S, Ahmad M. Recurrence relations for single and product moments of record values from size-biased Pareto distribution. Inter J Econ Busin Rev. 2015;3:139–144.
- Balakrishnan N, Balasubramanian K. A characterization of geometric distribution based on record values. J Appl Stat Sci. 1995;2:227–282.
- Franco M, Ruiz JM. On characterization of distributions by expected values of order statistics and record values with gap. Metrika. 1997;45:107–119. doi: 10.1007/BF02717097
- Franco M, Ruiz JM. Characterizations of discrete distributions based on conditional expectations of record values. Stat Papers. 2001;42:101–110. doi: 10.1007/s003620000043
- Wu JW. Characterizations of generalized mixtures of geometric and exponential distributions based on upper record values. Stat Papers. 2000;42:123–133. doi: 10.1007/s003620000045
- Wu JW. Characterizations of the finite mixture of exponential distributions by conditional moments of nonadjacent record values. J Japan Stat Soc. 2001;30:96–105.
- Bairamov I. On the characteristic properties of exponential distribution. Ann Inst Stat Math. 2000;52:448–458. doi: 10.1023/A:1004109117386
- Ahsanullah M, Shakil M, Kibria BMG. A characterization of power function distribution based on lower record values. Prob Stat Forum. 2013;6:68–72.
- Hassan AS, Nagy HF, Muhammed HZ, et al. Estimation of multicomponent stress-strength reliability following Weibull distribution based on upper record values. J Taib Univ Sci. 2020;14:244–253. doi: 10.1080/16583655.2020.1721751
- AL-Hussaini EK. Predicting observables from a general class of distributions. J Stat Plan Inf. 1999;79:79–91. doi: 10.1016/S0378-3758(98)00228-6
- Arnold BC, Balakrishnan N, Nagaraja HN. Records. New York: Wiley; 1998.
- Hwang JS, Lin GD. Extensions of Müntz-Szász theorem and applications. Analysis. 1984;4:143–160. doi: 10.1524/anly.1984.4.12.143
- Balakrishnan N, Chan PS. Record values from Rayleigh and Weibull distributions and associated inference. Proc Conf Ext Value Theo Appl. 1994;3:1471–1482.
- Balakrishnan N, Ahsanullah M. Relations for single and product moments of record values from Lomax distribution. Sankhy B. 1994;56:140–146.