MSO205A PRACTICE PROBLEMS SET 11 SOLUTIONS

Question 1. Let Xi ∼ P oisson(λi ), i = 1, 2, · · · , n be independent RVs, with λi > 0, ∀i. Show that
Pn
X1 + X2 + · · · + Xn ∼ P oisson( i=1 λi ).
(Note: A special case of this result is the following: If X1 , X2 , · · · , Xn be a random sample from
P oisson(λ) distribution, then X1 + X2 + · · · + Xn ∼ P oisson(nλ).)

Answer: Note that the MGF MXi (t) = exp(λi (et − 1)), ∀t ∈ R. Using independence of Xi ’s, we
have !
n
Y n
X
t
MX1 +X2 +···+Xn (t) = MXi (t) = exp λi (e − 1) , ∀t ∈ R.
i=1 i=1
Since the MGF, if it exists, determines the distribution, we conclude X1 + X2 + · · · + Xn ∼
P oisson(λ1 + λ2 + · · · + λn ).

Question 2. Let X ∼ P oisson(λ), Y ∼ P oisson(µ) be independent RVs. Find the conditional

distribution of X given X + Y = k for k = 0, 1, · · · .

Answer: We have X + Y ∼ P oisson(λ + µ) (by problem 2 above). Then, for k = 0, 1, · · ·

P(X = x and X + Y = k)
P(X = x|X + Y = k) =
P(X + Y = k)
P(X = x and Y = k − x)
=
P(X + Y = k)
P(X = x)P(Y = k − x)
= , (using independence of X and Y )
P(X + Y = k)
 −λ λx −µ µk−x
 e x! e (k−x)!

k , if x ∈ {0, 1, · · · , k},
(λ+µ)
= e−(λ+µ) k!

0, otherwise.
 x  k−x


 k
 λ
1− λ
, if x ∈ {0, 1, · · · , k},
x λ+µ λ+µ
=
0, otherwise.


λ
Therefore, X | X + Y = k ∼ Binomial(k, λ+µ ).

Question 3. Let X, Y be RVs defined on the same probability space. Fix a, b, c, d ∈ R and set
U = a + bX, V = c + dY . Express ρ(U, V ) in terms of ρ(X, Y ).

Answer: We have,

Cov(U, V ) = Cov(a + bX, c + dY ) = E[(a + bX)(c + dY )] − E(a + bX)E(c + dY ) = bdCov(X, Y )

and
V ar(U ) = E[(a + bX) − (a + bEX)]2 = b2 V ar(X), V ar(V ) = d2 V ar(Y ).
1
2 MSO205A PRACTICE PROBLEMS SET 11 SOLUTIONS

If b = 0 or d = 0, ρ(U, V ) is not defined. If bd 6= 0, then

Cov(U, V ) bd
ρ(U, V ) = p = ρ(X, Y ).
V ar(U )V ar(V ) |bd|

Question 4. Compute the factorial moments for Negative Binomial and Hypergeometric distribu-
tion.

Answer: Suppose X follows Negative Binomial(r, p) distribution. Then for any non-negative
integer k,

EX(X − 1) · · · (X − k + 1)
∞ 
X j+r−1 r
= j(j − 1) · · · (j − k + 1) p (1 − p)j
j=0
j
∞  
X j+r−1 r
= j(j − 1) · · · (j − k + 1) p (1 − p)j
j
j=k
∞  
X j+r−1
= pr (1 − p)k r(r + 1) · · · (r + k − 1) (1 − p)j−k
j−k
j=k
∞  
X j+k+r−1
= pr (1 − p)k r(r + 1) · · · (r + k − 1) (1 − p)j
j=0
j

= r(r + 1) · · · (r + k − 1)p−k (1 − p)k .

Suppose X follows the Hypergeometric distribution with parameters N, M and n. Then for any
non-negative integer k,

EX(X − 1) · · · (X − k + 1)
min{n,M } M

N −M


j n−j
X
= j(j − 1) · · · (j − k + 1) N


j=max{0,n−(N −M )} n

The above factorial moment is zero for all k > min{n, M }. For 1 ≤ k ≤ min{n, M },

EX(X − 1) · · · (X − k + 1)
min{n,M }   
M (M − 1) · · · (M − k + 1) X M −k N −M
= N j−k n−j


n j=max{k,n−(N −M )}
 
M (M − 1) · · · (M − k + 1) N − k
= N n−k


n

Question 5. Suppose a pair of fair die are rolled seven times independently. Find the probability
that the sum of the dots obtained is 12 once and 8 twice.
 MSO205A PRACTICE PROBLEMS SET 11 SOLUTIONS 3

Answer: Let X1 and X2 denote the number of times 12 and 8 appear in the seven rolls, respec-
tively.
1
Since the die are fair, the probability that 12 appears in a roll is 36 (the only favourable event being
5
(6, 6)) and the corresponding probability for 8 is 36 (favourable events being (2, 6), (3, 5), (4, 4), (5, 3), (6, 2)).
1 5 36−1−5
Hence, (X1 , X2 ) follows the Multinomial distribution with parameters 7 and 36 , 36 , 36 . There-
fore, the required probability is
1  2  4
5 × 6 × 7 56 7 57

7! 1 5 5
P(X1 = 1, X2 = 2) = = = .
1!2!4! 36 36 6 2 68 2 67

Question 6. If X1 , X2 , · · · , Xn are i.i.d. Geometric(p) RVs, for some p > 0, then find the distribu-
tion of X1 + X2 + · · · + Xn .

Answer: First consider the case n = 2. Recall that Geometric(p) distribution is the same as the
negative Binomial(1, p) distribution.
Consider any two independent RVs X and Y with distributions negative Binomial(m, p) and nega-
tive Binomial(n, p) respectively. Since the support of X and Y are exactly the set of non-negative
integers, X + Y is also discrete with the support contained in the set of non-negative integers. Now
for any non-negative integer k, using the independence of X and Y , we have
k
X
P(X + Y = k) = P(X = j, Y = k − j)
j=0
k
X
= P(X = j)P(Y = k − j)
j=0
k   
X j+m−1 k−j+n−1
= pm+n (1 − p)k
j=0
j k−j
 
k + m + n − 1 m+n
= p (1 − p)k ,
k
i.e. X + Y follows the negative Binomial(m + n, p) distribution. In the setting of the hypothesis,
X1 + X2 follows negative Binomial(2, p) distribution.
Consider the case n = 3. Here, X1 + X2 and X3 are independent. Hence, using the above case
X1 + X2 + X3 follows the negative Binomial(3, p) distribution.
By the principle of Mathematical Induction, X1 + X2 + · · · + Xn follows the negative Binomial(n, p)
distribution.

Question 7. Let X be a continuous RV with p.d.f. fX . If X is symmetric about µ ∈ R and if EX

exists, show that
z0.25 + z0.75
EX = µ = m = ,
2
where m, z0.25 , z0.75 denotes the median, the lower and upper quartiles respectively. Assume that
these are unique.
4 MSO205A PRACTICE PROBLEMS SET 11 SOLUTIONS

d
Answer: Let fX and FX denote the p.d.f. and the DF of X, respectively. Since, X − µ = µ − X,
we have E(X − µ) = E(µ − X), which implies EX = µ.
Now, FX−µ (x) = Fµ−X (x), ∀x ∈ R. But, FX−µ (x) = P(X − µ ≤ x) = P(X ≤ µ + x) and
Fµ−X (x) = P(µ − X ≤ x) = P(X ≥ µ − x). In particular, putting x = 0, we have P(X ≤ µ) =
P(X ≥ µ) = 1 − P(X < µ) = 1 − P(X ≤ µ), since, P(X = µ) = 0 by continuity of FX and therefore,
FX (µ) = 21 . Hence, m = µ.
Now, P(X ≤ z0.75 ) = 0.75 gives P(X − µ ≤ z0.75 − µ) = 0.75 or P(µ − X ≤ z0.75 − µ) = 0.75,
d
since X − µ = µ − X. Therefore, P(X ≥ 2µ − z0.75 ) = 0.75. Using the continuity of FX , P(X ≤
2µ − z0.75 ) = 1 − 0.75 = 0.25. Therefore, z0.25 = 2µ − z0.75 , which completes the proof.

Question 8. Let X be an RV with E|X| < ∞. Consider the function g : R → R defined by

g(x) := E|X − x|, x ∈ R. Show that g(m) ≤ g(x), ∀x ∈ R, where m is the median of X. (Note:
This shows that the mean deviation is minimized at the median).
Rm R∞
Answer: Let X be a continuous RV with p.d.f. fX . Note that −∞
fX (t) dt = m
fX (t) dt = 12 . If
x < m, then

g(x) − g(m)
Z ∞ Z ∞
= |t − x|fX (t) dt − |t − m|fX (t) dt
−∞ −∞
Z x Z m Z ∞ Z m Z ∞
= (x − t)fX (t) dt + (t − x)fX (t) dt + (t − x)fX (t) dt − (m − t)fX (t) dt − (t − m)fX (t) dt
−∞ x m −∞ m
Zx Z m Z ∞
= (x − m)fX (t) dt + (2t − x − m)fX (t) dt + (m − x)fX (t) dt
−∞ x m

≥ (m − x)[P(X ≥ m) − P(X ≤ x)] + (2x − x − m)P(x ≤ X ≤ m)

= (m − x)[P(X ≥ m) − P(X ≤ m)]

= 0.

A similar argument shows g(x) ≥ g(m) if x > m. This proves the case when X is continuous with
a p.d.f. fX .
The proof for the discrete case goes in an analogus manner.

X X
Question 9. Let X and Y be i.i.d. N (0, 1) RVs. Identify the distribution of Y and |Y | .

Answer: Since, Y 2 ∼ χ21 , using the independence of X and Y 2 , we have

X X
=q ∼ t1 .
|Y | Y2
1

Note that P(Y = 0) = 0. For z ∈ R,

         
X X X X X
P ≤z =P ≤ z, Y > 0 +P ≤ z, Y < 0 = P ≤ z, Y > 0 +P − ≤ z, Y < 0
Y Y Y |Y | |Y |
 MSO205A PRACTICE PROBLEMS SET 11 SOLUTIONS 5

d
Using the symmetry (X, Y ) = (−X, Y ),
   
X X
P − ≤ z, Y < 0 = P ≤ z, Y < 0
|Y | |Y |
and hence
       
X X X X
F X (z) = P ≤z =P ≤ z, Y > 0 + P ≤ z, Y < 0 = P ≤z .
Y Y |Y | |Y | |Y |
X
So, Y ∼ t1 .

n
Question 10. Let X ∼ Fm,n . Identify the distribution of n+mX .

d n Y1
Answer: Let Y1 ∼ χ2m and Y2 ∼ χ2n be independent RVs. Then X = m Y2 ∼ Fm,n . Therefore,

n d Y2
= .
n + mX Y1 + Y2
Since Y1 ∼ Gamma( m n
2 , 2) and Y2 ∼ Gamma( 2 , 2) are independent, using Question 5 of Problem
set 10 we have
n d Y2 n m
= ∼ Beta , .
n + mX Y1 + Y2 2 2

Question 11. Let X and Y be i.i.d. Exponential(λ) RVs, for some λ > 0. Identify the distribution
X
of Y .

2X
Answer: Since X ∼ Exponential(λ) = Gamma(1, λ), we have λ ∼ Gamma(1, 2) = χ22 . Similarly,
2Y
λ ∼ χ22 . Then,
! !−1
2X 2Y
X λ λ
= ∼ F2,2 .
Y 2 2

Question 12. Verify that for a discrete RV X with the DF





 0, if x < 0,

FX (x) := 12 , if 0 ≤ x < 1,



1, if x ≥ 1,


the median is not unique. Given p ∈ (0, 1), construct an example of discrete RV X (by specifying
the DF FX or the p.m.f. fX ) such that the quantile of order p is not unique.

1
Answer: For any x ∈ (0, 1), we have FX (x) = 2 and P(X = x) = 0. Hence the inequality
1
2 ≤ FX (x) ≤ 12 + P(X = x) is satisfied. Therefore, any x ∈ (0, 1) is a median for X.
Moreover, 21 = FX (0) ≤ 12 + P(X = 0) and hence 0 is also a median for X. Therefore, the median
is not unique in this case.
6 MSO205A PRACTICE PROBLEMS SET 11 SOLUTIONS

Given p ∈ (0, 1), consider the RV Y given by the DF FY (or equivalently, the p.m.f. fY )
 



 0, if x < 0, 


 p, if x = 0,
 
FY (x) := p, if 0 ≤ x < 1, , fY (x) = 1 − p, if x = 1,

 

 
1, if x ≥ 1. 0, otherwise.

 


Similar to above argument for X, the quantile of order p for Y is not unique.

Question 13. Verify that for a continuous RV X with the DF




 0, if x < 0,


 x , if 0 ≤ x < 1,


2


FX (x) := 1
 2 , if 1 ≤ x < 2,


x−1
2 , if 2 ≤ x < 3,







1, if x ≥ 3,


the median is not unique. Given p ∈ (0, 1), construct an example of continuous RV X (by specifying
the DF FX or the p.d.f. fX ) such that the quantile of order p is not unique.

1
Answer: For a continuous RV X, a median x is a solution to the equation FX (x) = 2. In this
case, this equation is solved by all x ∈ [1, 2]. Hence, the median is not unique in this case.
Given p ∈ (0, 1), consider the RV Y given by the DF FY (or equivalently, the p.d.f. fY )




 0, if x < 0,

 




 xp, if 0 ≤ x < 1, p, ∀x ∈ [0, 1)



 
FY (x) = p, if 1 ≤ x < 2, , fY (x) = 1 − p, ∀x ∈ [2, 3)

 

 
p(3 − x) + x − 2, if 2 ≤ x < 3, 0, otherwise.

 






1, if x ≥ 3.


Similar to above argument for X, the quantile of order p for Y is not unique.

Question 14. Consider the set

n  Pp  o
A := t = (t1 , t2 , . . . , tp ) ∈ Rp : E e i=1 ti Xi < ∞

for a given random vector X = (X1 , X2 , · · · , Xp ) and look at ΨX (t) := ln MX (t), t ∈ A. Verify that

∂2
 
ΨX (t) = Cov(Xi , Xj ).
∂ti ∂tj (t1 ,t2 ...tp )=(0,...,0)
 MSO205A PRACTICE PROBLEMS SET 11 SOLUTIONS 7

Answer: For i 6= j with i, j ∈ {1, . . . , p}, we have

Cov (Xi , Xj )

= E (Xi Xj ) − E (Xi ) E (Xj )

∂2
     
∂ ∂
= MX (t) − MX (t) MX (t)
∂ti ∂tj (t1 ,t2 ...tp )=(0,...,0) ∂ti (t1 ,t2 ...tp )=(0,...,0) ∂tj (t1 ,t2 ...tp )=(0,...,0)
2
 
∂
= ΨX (t) ,
∂ti ∂tj (t1 ,t2 ...tp )=(0,...,0)

where the last equality follows from the one-dimensional case.

h 2 i

2 ∂
When i = j, we have Cov (Xi , Xj ) = V ar(Xi ) = E Xi2 − (EXi ) = ∂t 2 ΨX (t) ,
i (t1 ,t2 ...tp )=(0,...,0)
also similar to the one-dimensional case.