PSLecture15 2022

Proability and Statistics
MA20205
Bibhas Adhikari
Autumn 2022-23, IIT Kharagpur
Lecture 15
October 25, 2022
Bibhas Adhikari (Autumn 2022-23, IIT Kharagpur) Proability and Statistics Lecture 15 October 25, 2022 1/8
Bivariate normal distribution
Introduced by Galton and Dickson in 1886.
Introduced by Galton and Dickson in 1886.A continuous bivariate random

variable (X , Y ) is said to have the bivariate normal distribution if the joint
pdf is of the form
1 1
f (x, y ) = p e − 2 Q(x,y )
2πσ1 σ2 1 − ρ2
where µ1 , µ2 ∈ R, σ1 , σ2 ∈ (0, ∞) and ρ ∈ (−1, 1) are parameters, and

" #
x − µ1 2 y − µ2 2

1 x − µ1 y − µ2
Q(x, y ) = − 2ρ +
1 − ρ2 σ1 σ1 σ2 σ2
Introduced by Galton and Dickson in 1886.A continuous bivariate random

variable (X , Y ) is said to have the bivariate normal distribution if the joint
pdf is of the form
1 1
f (x, y ) = p e − 2 Q(x,y )
2πσ1 σ2 1 − ρ2
where µ1 , µ2 ∈ R, σ1 , σ2 ∈ (0, ∞) and ρ ∈ (−1, 1) are parameters, and

" #
x − µ1 2 y − µ2 2

1 x − µ1 y − µ2
Q(x, y ) = − 2ρ +
1 − ρ2 σ1 σ1 σ2 σ2
Denote:
(X , Y ) ∼ N (µ1 , µ2 , σ1 , σ2 , ρ)
Bivariate normal/Gaussian distribution
Parameters:
1 The surface f (x, y ) has the shape of mountain
Parameters:
2 The pair (µ , µ ) gives the center of the mountain located in the

1 2
(x, y ) plane
Parameters:

1 2
(x, y ) plane
3 σ 2 and σ 2 measure the spread of the mountain in the x-direction and
1 2
y -direction respectively
Parameters:

1 2
(x, y ) plane
3 σ 2 and σ 2 measure the spread of the mountain in the x-direction and
1 2
y -direction respectively
4 ρ determines the shape and orientation
Watch: The link
Observation If we vertically slice the joint pdf of bivariate normal

distribution then we have a univariate normal distribution.

distribution then we have a univariate normal distribution. Thus the
marginal pdfs are also normal.

2
1 − 21
x−µ1
fX (x) = √ e σ1
σ1 2π

2
1 − 21
x−µ1
fX (x) = √ e σ1
σ1 2π
y −µ2 2

1 −1
fY (y ) = √ e 2 σ2
σ2 2π

2
1 − 21
x−µ1
fX (x) = √ e σ1
σ1 2π
y −µ2 2

1 −1
fY (y ) = √ e 2 σ2
σ2 2π
Remark However the converse need not be true!!
proof of marginal distribution
"
x − µ1 2

1 x − µ1 y − µ2
Q(x, y ) = − 2ρ
1 − ρ2 σ1 σ1 σ2
#
y − µ2 2

+
σ2
"
x − µ1 2

1 x − µ1 y − µ2
Q(x, y ) = − 2ρ
1 − ρ2 σ1 σ1 σ2
#
y − µ2 2

+
σ2
" 2 #
y − µ2 2

1 x − µ1 y − µ 2
= −ρ + (1 − ρ2 )
1 − ρ2 σ1 σ2 σ2
(x − a)2 (y − µ2 )2
= +
(1 − ρ2 )σ12 σ22
where a = µ1 + ρ σσ12 (y − µ2 ).
"
x − µ1 2

1 x − µ1 y − µ2
Q(x, y ) = − 2ρ
1 − ρ2 σ1 σ1 σ2
#
y − µ2 2

+
σ2
" 2 #
y − µ2 2

1 x − µ1 y − µ 2
= −ρ + (1 − ρ2 )
1 − ρ2 σ1 σ2 σ2
(x − a)2 (y − µ2 )2
= +
(1 − ρ2 )σ12 σ22
where a = µ1 + ρ σσ12 (y − µ2 ).Hence
Z ∞ (y −µ2 )2 Z ∞ (x−a)2
1 −
2σ 2
−
2(1−ρ2 )σ 2
fY (y ) = f (x, y )dx = p e 2 e 1 dx
−∞ πσ1 σ2 1 − ρ2 −∞
If (X , Y ) ∼ N (µX , µY , σX , σY , ρ) then
E (X ) = µX , E (Y ) = µY , Var(X ) = σX2 , Var(Y ) = σY2

1 2 2 +2ρσ 2 2
Correlation coefficient = ρ, M(s, t) = e µX s+µY t+ 2 (σX s X σY st+σY t )

1 2 2 +2ρσ 2 2
Proof X ∼ N (µX , σX2 ), Y ∼ N (µY , σY2 ). Then

W = sX + tY ∼ N (µW , σW 2 ) where
2
µW = sµX + tµY , σW = s 2 σX2 + 2stρσX σY + t 2 σY .

1 2 2 +2ρσ 2 2

2
1 2 2
Therefore, the mgf of W is M(τ ) = e µW τ + 2 τ σW
.

1 2 2 +2ρσ 2 2

2
1 2 2
Therefore, the mgf of W is M(τ ) = e µW τ + 2 τ σW
.Then mgf of (X , Y ) is
1 2
M(s, t) = E (e sX +tY ) = e µW + 2 σW
1 2 2 +2ρσ 2 2
= e µX s+µY t+ 2 (σX s X σY st+σY t )
Let f (x, y ) be the joint pdf of (X , Y ). Then conditional density of Y given

X = x is 2
y −b
1 − 12 √
2
fY |X (y |x) = p e σY 1−ρ
σY 2π(1 − ρ2 )
where
σY
b = µY + ρ (x − µX ).
σX
Let f (x, y ) be the joint pdf of (X , Y ). Then conditional density of Y given

X = x is 2
y −b
1 − 12 √
2
fY |X (y |x) = p e σY 1−ρ
σY 2π(1 − ρ2 )
where
σY
b = µY + ρ (x − µX ).
σX
Similarly, 2
x−c
1 − 12 √
1−ρ2
fX |Y (x|y ) = p e σX
σX 2π(1 − ρ2 )
where
σX
c = µX + ρ (y − µY )
σY
σY
E (Y |x) = µY + ρ (x − µX )
σX
σY
E (Y |x) = µY + ρ (x − µX )
σX
σX
E (X |y ) = µX + ρ (y − µY )
σY
σY
E (Y |x) = µY + ρ (x − µX )
σX
σX
E (X |y ) = µX + ρ (y − µY )
σY
Var(Y |x) = σY (1 − ρ2 )
2
σY
E (Y |x) = µY + ρ (x − µX )
σX
σX
E (X |y ) = µX + ρ (y − µY )
σY
Var(Y |x) = σY (1 − ρ2 )
2
Var(X |y ) = σX2 (1 − ρ2 )
σY
E (Y |x) = µY + ρ (x − µX )
σX
σX
E (X |y ) = µX + ρ (y − µY )
σY
Var(Y |x) = σY (1 − ρ2 )
2
Var(X |y ) = σX2 (1 − ρ2 )
Interesting result (Cramer, 1941)

Two random variables X and Y have a joint bivariate normal distribution
if and only if every linear combination of X and Y has univariate normal
distribution

PSLecture15 2022

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PSLecture15 2022

Uploaded by

Copyright:

Available Formats

Proability and Statistics

Autumn 2022-23, IIT Kharagpur

Introduced by Galton and Dickson in 1886.

Introduced by Galton and Dickson in 1886.A continuous bivariate random

where µ1 , µ2 ∈ R, σ1 , σ2 ∈ (0, ∞) and ρ ∈ (−1, 1) are parameters, and

Introduced by Galton and Dickson in 1886.A continuous bivariate random

where µ1 , µ2 ∈ R, σ1 , σ2 ∈ (0, ∞) and ρ ∈ (−1, 1) are parameters, and

2 The pair (µ , µ ) gives the center of the mountain located in the

2 The pair (µ , µ ) gives the center of the mountain located in the

2 The pair (µ , µ ) gives the center of the mountain located in the

Watch: The link

Observation If we vertically slice the joint pdf of bivariate normal

Observation If we vertically slice the joint pdf of bivariate normal

Observation If we vertically slice the joint pdf of bivariate normal

Observation If we vertically slice the joint pdf of bivariate normal

Observation If we vertically slice the joint pdf of bivariate normal

E (X ) = µX , E (Y ) = µY , Var(X ) = σX2 , Var(Y ) = σY2

E (X ) = µX , E (Y ) = µY , Var(X ) = σX2 , Var(Y ) = σY2

E (X ) = µX , E (Y ) = µY , Var(X ) = σX2 , Var(Y ) = σY2

Proof X ∼ N (µX , σX2 ), Y ∼ N (µY , σY2 ). Then

E (X ) = µX , E (Y ) = µY , Var(X ) = σX2 , Var(Y ) = σY2

Proof X ∼ N (µX , σX2 ), Y ∼ N (µY , σY2 ). Then

E (X ) = µX , E (Y ) = µY , Var(X ) = σX2 , Var(Y ) = σY2

Proof X ∼ N (µX , σX2 ), Y ∼ N (µY , σY2 ). Then

Let f (x, y ) be the joint pdf of (X , Y ). Then conditional density of Y given

Let f (x, y ) be the joint pdf of (X , Y ). Then conditional density of Y given

Interesting result (Cramer, 1941)

You might also like