Order Statistics - Lecture 9
Order Statistics - Lecture 9
Asymptotic Results
of Extreme Order
Statistics
Prof. Magdy E.
Lecture 9: Some Asymptotic Results of
El-Adll
An Introduction
Extreme Order Statistics
Some Asymptotic
Results for
Exponential Order
Statistics
Prof. Magdy E. El-Adll
Asymptotic
Distribution of the
Sample Maximum
Department of Mathematics, Faculty of Science,
Helwan University, Ain Helwan, Cairo, Egypt.
22 / 4 / 2024
Prof. Magdy E. El-Adll ( Department of Mathematics, Faculty
Lecture of
9: Science,
Some Asymptotic
Helwan University,
Results ofAin
Extreme
Helwan,
Order
Cairo,
Statistics
Egypt.) 22 / 4 / 2024 1 / 34
Presentation Outline
Lecture 9: Some
Asymptotic Results
of Extreme Order
Statistics
Prof. Magdy E.
El-Adll 1 An Introduction
An Introduction
Some Asymptotic
Results for 2 Some Asymptotic Results for Exponential Order Statistics
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
3 Asymptotic Distribution of the Sample Maximum
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
of its special functions have received all of our attention
El-Adll
in the preceding chapters.
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
of its special functions have received all of our attention
El-Adll
in the preceding chapters.
An Introduction
An exception was when we considered approximations to
Some Asymptotic
Results for
Exponential Order
the moments of order statistics.
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
of its special functions have received all of our attention
El-Adll
in the preceding chapters.
An Introduction
An exception was when we considered approximations to
Some Asymptotic
Results for
Exponential Order
the moments of order statistics.
Statistics
Past experiences of the distribution theory of order s-
Asymptotic
Distribution of the
Sample Maximum
tatistics and their functions (such as quasi-ranges) for fi-
nite samples have repeatedly demonstrated that the exact
CDFs of such statistics are computationally messy, except
in rare cases.
Prof. Magdy E. El-Adll ( Department of Mathematics, Faculty
Lecture of
9: Science,
Some Asymptotic
Helwan University,
Results ofAin
Extreme
Helwan,
Order
Cairo,
Statistics
Egypt.) 22 / 4 / 2024 4 / 34
An introduction
Lecture 9: Some
Asymptotic Results
of Extreme Order
Statistics
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Some Asymptotic
of interest, the extreme-value theory (EVT) provides, in
Results for
Exponential Order some cases, exact probabilistic models but, in most cas-
Statistics
Asymptotic
es,approximate probabilistic ones.
Distribution of the
Sample Maximum
Some Asymptotic
of interest, the extreme-value theory (EVT) provides, in
Results for
Exponential Order some cases, exact probabilistic models but, in most cas-
Statistics
Asymptotic
es,approximate probabilistic ones.
Distribution of the
Sample Maximum A complicated situation can therefore be replaced by a
simple asymptotic model if the fundamental conditions of
the actual situation are compatible with the model as-
sumptions.
Prof. Magdy E. El-Adll ( Department of Mathematics, Faculty
Lecture of
9: Science,
Some Asymptotic
Helwan University,
Results ofAin
Extreme
Helwan,
Order
Cairo,
Statistics
Egypt.) 22 / 4 / 2024 5 / 34
An introduction-continued
Lecture 9: Some
Asymptotic Results
of Extreme Order With respect to the limiting behavior of a sequence of random
Statistics
Prof. Magdy E.
variables, there exist several mods of convergence.
El-Adll
Convergence in distribution or weak convergence,
An Introduction
Prof. Magdy E.
variables, there exist several mods of convergence.
El-Adll
Convergence in distribution or weak convergence,
An Introduction
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
As in the proceeding chapters, we start with a random
El-Adll
sample X1 , ..., Xn from a continuous population whose
An Introduction CDF is F .
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
As in the proceeding chapters, we start with a random
El-Adll
sample X1 , ..., Xn from a continuous population whose
An Introduction CDF is F .
Some Asymptotic
Results for
Exponential Order
The asymptotic behavior of a single-order statistic Xr:n
Statistics
Asymptotic
depends on:
Distribution of the
Sample Maximum
Prof. Magdy E.
As in the proceeding chapters, we start with a random
El-Adll
sample X1 , ..., Xn from a continuous population whose
An Introduction CDF is F .
Some Asymptotic
Results for
Exponential Order
The asymptotic behavior of a single-order statistic Xr:n
Statistics
Asymptotic
depends on:
Distribution of the
Sample Maximum (1) How r relates to n.
Prof. Magdy E.
As in the proceeding chapters, we start with a random
El-Adll
sample X1 , ..., Xn from a continuous population whose
An Introduction CDF is F .
Some Asymptotic
Results for
Exponential Order
The asymptotic behavior of a single-order statistic Xr:n
Statistics
Asymptotic
depends on:
Distribution of the
Sample Maximum (1) How r relates to n.
(2) The functional form of F .
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Some Asymptotic Central order statistic when r = [np] + 1 with 0 < p < 1,
Results for
Exponential Order
Statistics
where[.] stands for the greatest integer part.
Asymptotic
Distribution of the
Sample Maximum
Some Asymptotic Central order statistic when r = [np] + 1 with 0 < p < 1,
Results for
Exponential Order
Statistics
where[.] stands for the greatest integer part.
Asymptotic Intermediate order statistic when both r and n − r
Distribution of the r
Sample Maximum
approach infinity, but → 0 or 1.
n
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll We focus on the extreme case; more specifically, we con-
An Introduction sider the sample maximum and minimum.
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll We focus on the extreme case; more specifically, we con-
An Introduction sider the sample maximum and minimum.
Some Asymptotic
Results for
The most important message about extreme order statis-
Exponential Order
Statistics
tics is that:
Asymptotic If the limit distribution exists it is nonnormal
Distribution of the
Sample Maximum Depends on F only through its tail behavior.
Prof. Magdy E.
El-Adll 1 An Introduction
An Introduction
Some Asymptotic
Results for 2 Some Asymptotic Results for Exponential Order Statistics
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
3 Asymptotic Distribution of the Sample Maximum
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll
Before studying the three possible limit distributions for
An Introduction
order statistics from an arbitrary CDF, we first consider
Some Asymptotic
Results for
Exponential Order
the asymptotic results for the exponential distribution.
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll
Before studying the three possible limit distributions for
An Introduction
order statistics from an arbitrary CDF, we first consider
Some Asymptotic
Results for
Exponential Order
the asymptotic results for the exponential distribution.
Statistics
Asymptotic
These results, while easy to prove, give us an initial idea
Distribution of the
Sample Maximum
about the nature of the limit distributions.
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
Without any loss of generality, we assume the scale param-
El-Adll
eter θ = 1 and recall once again the ever useful
An Introduction
representation (3.5) for exponential order statistics given
Some Asymptotic
Results for
Exponential Order
in Theorem (3.1).
Statistics
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
Without any loss of generality, we assume the scale param-
El-Adll
eter θ = 1 and recall once again the ever useful
An Introduction
representation (3.5) for exponential order statistics given
Some Asymptotic
Results for
Exponential Order
in Theorem (3.1).
Statistics
Asymptotic
It expresses Xr:n as a linear function of independent and
Distribution of the
Sample Maximum identically distributed standard exponential random vari-
ables denoted by Zi ’s as follows:
Prof. Magdy E.
have where
El-Adll
d
Z1 d Z ,
An Introduction
X1:n = or equivalently nX1:n = 1
Some Asymptotic
Results for
n
Exponential Order
Statistics
Asymptotic
which implies
Distribution of the
Sample Maximum
d Z
nX1:n −→ as n → ∞.
1
Asymptotic
Therefore, we have
Distribution of the
Sample Maximum
d
nZ1 nZ2 nZr
nXr:n = + + ··· + .
n n−1 n−r+1
An Introduction d
nXr:n −→ {Z1 + Z2 + · · · + Zr } as n → ∞.
Some Asymptotic
Results for
Exponential Order
Statistics That is, the limiting distribution of nXr:n is gamma
Asymptotic
Distribution of the distribution with scale parameter one and shape parameter of
Sample Maximum
r, where r is any fixed integer.
Prof. Magdy E.
with the CDF for the sample maximum. It can be noted that,
El-Adll
Asymptotic
Distribution of the
Sample Maximum 0, if x ≤ − log n;
= n
{1 − exp [−(x + log n)]} , if x > − log n.
Prof. Magdy E.
P (Xn:n − log n ≤ x)
El-Adll
An Introduction
0, n if x ≤ − log n;
= 1
Some Asymptotic
Results for 1 − exp(−x) , if x > − log n.
Exponential Order n
Statistics
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Asymptotic
Distribution of the
Sample Maximum
Asymptotic
butions.
Distribution of the
Sample Maximum
Asymptotic
butions.
Distribution of the
Sample Maximum In the extreme case, the limit law depends on r.
We will build on these findings in the next sections, assuming
F is an arbitrary absolutely continuous function.
Prof. Magdy E. El-Adll ( Department of Mathematics, Faculty
Lecture of
9: Science,
Some Asymptotic
Helwan University,
Results ofAin
Extreme
Helwan,
Order
Cairo,
Statistics
Egypt.) 22 / 4 / 2024 23 / 34
Presentation Outline
Lecture 9: Some
Asymptotic Results
of Extreme Order
Statistics
Prof. Magdy E.
El-Adll 1 An Introduction
An Introduction
Some Asymptotic
Results for 2 Some Asymptotic Results for Exponential Order Statistics
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
3 Asymptotic Distribution of the Sample Maximum
An Introduction
Fn:n (x) = P (Xn:n ≤ x) = F n (x), p
Xn:n −→ F −1 (1),
Some Asymptotic
Results for
Exponential Order which is the upper limit of the support of F (sometimes called
Statistics
Asymptotic
the right-end point of F ). Recall that a CDF H is called
Distribution of the
Sample Maximum degenerate if
0, if; x < α
H(x) = (2)
1, if; x ≥ α,
Prof. Magdy E. El-Adll ( Department of Mathematics, Faculty
Lecture of
9: Science,
Some Asymptotic
Helwan University,
Results ofAin
Extreme
Helwan,
Order
Cairo,
Statistics
Egypt.) 22 / 4 / 2024 25 / 34
The normalized maximum
Lecture 9: Some
Asymptotic Results
of Extreme Order
Statistics
In order to hope for a nondegenerate limit distribution, we will
Prof. Magdy E.
El-Adll have to appropriately normalize or standardize
Xn:n .
An Introduction Xn:n − an
Some Asymptotic
In other words, we look at the sequence , n ≥ 1,
Results for bn
Exponential Order
Statistics where an represents a shift in location and bn > 0 represents
Asymptotic
Distribution of the a change in scale. In this case, the CDF of the normalized
maximum is given by F n (an + bn x).
Sample Maximum
Prof. Magdy E.
El-Adll
An Introduction
Some Asymptotic
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Some Asymptotic
points of a nondegenerate CDF G?
Results for
Exponential Order
Statistics
Asymptotic
Distribution of the
Sample Maximum
Some Asymptotic
points of a nondegenerate CDF G?
Results for
Exponential Order 2 What kind of CDF G can appear as the limiting CDF?
Statistics
Asymptotic
Distribution of the
Sample Maximum
Some Asymptotic
points of a nondegenerate CDF G?
Results for
Exponential Order 2 What kind of CDF G can appear as the limiting CDF?
Statistics
Some Asymptotic
points of a nondegenerate CDF G?
Results for
Exponential Order 2 What kind of CDF G can appear as the limiting CDF?
Statistics
Prof. Magdy E.
El-Adll
Asymptotic
Distribution of the
Sample Maximum
Prof. Magdy E.
A CDF F is said to belong to the max domain of attraction
El-Adll
of a nondegenerate CDF G if there exist sequences {an } and
An Introduction
Some Asymptotic
{bn > 0} such that
Results for
Exponential Order
Statistics
lim F n (an + bn x) = G(x) (3)
Asymptotic n→∞
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll
Let W be a random variable whose CDF is G. Then, asserting
An Introduction
that F ∈ D(G) is equivalent to saying
Some Asymptotic
Results for Xn:n − an d
−→ W.
Exponential Order
Statistics
Asymptotic
bn
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll
Let W be a random variable whose CDF is G. Then, asserting
An Introduction
that F ∈ D(G) is equivalent to saying
Some Asymptotic
Results for Xn:n − an d
−→ W.
Exponential Order
Statistics
Asymptotic
bn
Distribution of the
Sample Maximum
Prof. Magdy E.
El-Adll Definition 2
An Introduction Two distribution functions F1 and F2 are said to be of the same
Some Asymptotic
Results for
type if there exist constants a0 and b0 > 0 such that
Exponential Order
Statistics
Asymptotic
Distribution of the
F1 (a0 + b0 x) = F2 (x).
Sample Maximum
Some Asymptotic
A cumulative distribution function F cannot be in the max
Results for
Exponential Order domain of attraction for more than one type.
Statistics
Asymptotic
Distribution of the
Sample Maximum
for some constants a0r and b0r > 0. Thus, G must be such that