RS – 4 – Jointly Distributed RV (a)
Chapter 4 Jointly distributed Random variables
Continuous Multivariate distributions
Continuous Random Variables
1 RS – 4 – Jointly Distributed RV (a)
Joint Probability Density Function (pdf)
Definition: Joint Probability density function Two random variable are said to have joint probability density function f(x,y) if
1. 0 f xy ,
2. f x , y dxdy 1
3. P X , Y A f x , y dxdy A
Marginal and Condition Density
Definition: Marginal Density Let X and Y denote two RVs with joint pdf f(x,y), then the marginal density of X is f xfxydy , X f yfxydx , and the marginal density of Y is Y
Definition: Conditional Density Let X and Y denote two RVs with joint pdf f(x,y) and marginal
densities fX(x), fY(y), then the conditional density of Y given X = x and the conditional density of X given Y = y are given by f xy, f xy, fyxYX fxyXY f X x f Y y
2 RS – 4 – Jointly Distributed RV (a)
Joint MGF Definition: MGF of (X,Y) Let X and Y be two RVs with joint pdf f(x,y) then the MGF of X Y: m XY (t1,t2 ) E[exp(t1 X t2Y )] exp(t1 X t2Y ) f (x, y)dxdy R 2 Theorem: The MGF of a pair of independent RVs is the product of the MGF of the corresponding marginal distributions. That is,
mXY(t1,t2) = mX(t1) mY(t2). Proof: m XY (t1 , t 2 ) exp( t1 X t 2Y ) f (x, y)dxdy exp( t1 X ) exp( t 2Y ) f (x) f ( y)dxdy exp( t1 X ) f (x)dx exp( t 2Y ) f ( y)dy m X (t1 )mY (t 2 )
Marginal MGF
Definition: MGF of the marginal distribution of X (andY)
Let mXY(t1,t2) be the MGF of (X,Y), then the MGF of the marginal distributions of X and Y are, respectively, mXY(t1,0) and mXY(0,t2).
Proof: m X (t) exp(tX ) f X (x)dx exp(tX )[ f XY (x, y)dy]dx
exp(tX ) f XY (x, y)dydx m XY (t,0) Similar derivation for Y.
3 RS – 4 – Jointly Distributed RV (a)
The bivariate Normal distribution
Sir Francis Galton (1822 –1911, England)
The bivariate normal distribution
Let the joint distribution be given by: 1 1 Qx12, x fxx, e2 12 2 2112 where 2 2 xxxx 112 11 2 2 2 2 1122 Qxx, 12 1 2
This distribution is called the bivariate Normal distribution.
The parameters are 1, 2 , 1, 2 and The properties of this distribution were studied by Francis Galton and discovered its relation to the regression, term Galton coined.
4 RS – 4 – Jointly Distributed RV (a)
The bivariate normal distribution Surface Plots of the bivariate Normal distribution
The bivariate normal distribution Note: We can have a more compact joint using linear algebra:
2 1 1 x x x x f (x , x ) exp ( 1 1 )2 2( 1 1 )( 2 2 ) ( 2 2 )2 1 2 2 2(1 2 ) 21 2 (1 ) 1 1 2 2 1 1 exp (x μ)'1(x μ) 2(| |)1/2 2
(1) Determine the inverse and determinant of Σ (the covariance matrix)
2 2 1 21 2 2 2 2 2 12 2 2 2 2 | |1 2 12 1 2 (1 2 2 ) 1 2 (1 ) 12 2 1 2 2 1 1 2 21 2 | | 12 1
5 RS – 4 – Jointly Distributed RV (a)
The bivariate normal distribution
(2) Write a quadratic form for Q(x1,x2):
2 2 xxxx 112 11 2 2 2 2 1122 Qxx, 12 1 2
1 Q(x1, x2 ) (x μ)' (x μ) 2 1 2 21 x1 1 (x1 1) (x2 2 ) 2 | | 12 1 x2 2
1 2 2 x1 1 (x1 1) 2 (x2 2 )12 (x1 1) 21 (x2 2 )1 | | x2 2 1 ((x )2 2 2 (x )(x ) (x )2 2 ) | | 1 1 2 21 1 1 2 2 2 2 1 2 2 2 2 ((x1 1) 2 2 21(x1 1)(x2 2 ) (x2 2 ) 1 ) 2 2 2 1 2 (1 )
The Bivariate Normal Distribution
f(x,y)
12 12 y 12 y y
x x x Contour Plots of the Bivariate Normal Distribution
yyy 12 12 12
2 2 2
x xx 1 1 1 Scatter Plots of data from the Bivariate Normal Distribution
yyy 12 12 12
2 2 2
x xx 1 1 1
6 RS – 4 – Jointly Distributed RV (a)
The bivariate normal distribution: MGF
Let MGF of a bivariate normal is given by: 1 m (t ,t ) exp[t t (t 2 2 t 2 2 2 t t )] XY 1 2 1 X 2 Y 2 1 X 2 Y XY 1 2 X Y
Note: When ρXY = 0 –i.e., X and Y are independent. The MGF is: 1 m (t ,t ) exp[t t (t 2 2 t 2 2 )] XY 1 2 1 X 2 Y 2 1 X 2 Y
Marginal distributions for the Bivariate Normal
Recall the definition of marginal distributions for continuous RV: f xfxxdx , and f xfxxdx , 11 12 2 22 12 1 In the case of the bivariate normal distribution the marginal
distribution of xi is Normal with mean i and standard deviation i. Proof:
The marginal distributions of x2 is 1 1 Qx12, x f xfxxdx , e2 dx 22 12 1 2 1 2112 where 22 xxxx 112 11 2 2 2 2 1122 Qxx, 12 1 2
7 RS – 4 – Jointly Distributed RV (a)
Marginal distributions for the Bivariate Normal
Now: 22 xxxx 112 11 2 2 2 2 1122 Qxx, 12 1 2
2 22 xa11 x aa cxc222 1 2 bbbb xx2 11222 x 1122 1 21 21
2 2 xx 1 222 22 + 22 21 22 1212111
Marginal distributions for the Bivariate Normal
22 2 Hence bb 1 or 1 1
a x 122 Also b2 22 1121
1 1 12x 2 1 2 and 1 ax12 2
8 RS – 4 – Jointly Distributed RV (a)
Marginal distributions for the Bivariate Normal
Finally 2 a 2 2 xx c 1 222 22 b 2 22 21 22 1212111
2 2 xxa 2 c 1 2 22 22 22 21 22b 2 1212111
2 2 xx 1 222 22 22 21 22 1212111 2 1 x 12 2 2 1
Marginal distributions for the Bivariate Normal
and 2 1 2 112 cxx12 122 2 22 22 1 1 22 2 1 122x 2 2 1 1 2 2 2 1 x22 22 1 1 2
2 x 22 2
9 RS – 4 – Jointly Distributed RV (a)
Marginal distributions for the Bivariate Normal
Summarizing 22 xxxx 112 11 2 2 2 2 1122 Qxx, 12 1 2 2 xa1 c b
2 where b 1 1
1 ax12 2 2 x and c 22 2
Marginal distributions for the Bivariate Normal
f xfxxdx , Thus 22 12 1 1 1 Qx12, x e2 dx 2 1 2112
2 1 xa1 c 1 2 b edx 2 1 2112
2 c 2 1 xa1 21 be e2 b dx 2 2 b 1 2112
2 1 x 1 22 e 2 2 22
10 RS – 4 – Jointly Distributed RV (a)
Marginal distributions for the Bivariate Normal
• Thus the marginal distribution of x2 is Normal with mean 2 and standard deviation 2.
• Similarly, the marginal distribution of x1 is Normal with mean 1 and standard deviation 1.
Note: This derivation is much easier using MGFs. Use the MGF of a bivariate normal. To get the MGF of the marginal of
X, set t2=0. 1 m (t ,t ) exp[t t (t 2 2 t 2 2 2 t t )] XY 1 2 1 X 2 Y 2 1 X 2 Y XY 1 2 X Y 1 m (t ,0) exp[t (t 2 2 )] m (t ) XY 1 1 X 2 1 X X 1
Marginal distributions: Bivariate Normal
Bivariate Normal Distribution with marginal distributions
11 RS – 4 – Jointly Distributed RV (a)
Conditional distributions for the Bivariate Normal
Recall the definition of conditional distributions for continuous RVs:
f xx12, f xx12, fxx12 12 and fxx21 21 f22x f11x In the case of the bivariate normal distribution the conditional
distribution of xi given xj is Normal with mean and standard deviation: i 2 ij ijjx and ij i 1 j
Conditional distributions: Bivariate Normal
1 Proof: | Qx12, x e 2 2 f (x1, x2 ) 2112 f 2 1|2 1 x 22 f (x2 ) 1 e 2 2 22
2 2 2 11x 11xa122 x Qx, x | 22 c | 12 22b ee22 2 2 22 2111 21 where 2 x 2 a 1 (x ) c 22 b 1 1 1 2 2 2 2
12 RS – 4 – Jointly Distributed RV (a)
Conditional distributions: Bivariate Normal
2 1xa1 1 2b Hence fxx12 12 e 2 b
Then, the conditional distribution of x2 given x1 is Normal with mean and standard deviation:
1 2 ax12 122 and b 12 1 1 2
Conditional distributions: Bivariate Normal
• Bivariate Normal Distribution with conditional distribution
• Using matrix notation, the conditional moments are given by: 1 1|2 1 12 22 ( x2 2 ) 1 1|2 11 12 22 21
13 RS – 4 – Jointly Distributed RV (a)
Conditional distributions: Bivariate Normal
Major axis of ellipses x2
( 1, 2)
2 2|1 2 (x1 1 ) 1
Regression to the mean (μ1= μ2= μ) :
2|1 (x1 ); 0 1
x1 ( 12 ) 2 (x ) 12 (x ) Note: 2|1 2 1 1 2 2 1 1 12 1 1
Conditional distributions: KF Application
• The Kalman filter (KF) uses the observed data to learn about the unobservable state variables, which describe the state of the model.
• KF models dynamically what we measure, zt, and the state, yt. In the simple, linear model we have:
yt = Ayt-1 + wt (state or transition equation)
zt = Hyt + vt (measurement equation)
wt, vt: error terms, with zero mean and variance Q and R, respectively.
• Based on time t-1 information, the KF generates predictions for yt: yt|t-1 = A yt-1|t-1 + B ut T Pt|t-1 = A Pt-1 A + Q (conditional variance of yt ) • It also generates an update, once the information t is known: y = y + P HT (F )-1 e t|t t|t-1 t|t-1 t|t-1 t|t-1 28 T -1 Pt|t = Pt|t-1 – Pt|t-1 H (Ft|t-1) HPt|t-1
14 RS – 4 – Jointly Distributed RV (a)
Conditional distributions: KF Application
• We define the forecast error for the observed zt and its variance as: et|t-1 = zt –zt|t-1 = zt – Hyt|t-1 T Ft|t-1= H Pt|t-1 H + R
Then, we write the joint of distribution of (yt , et)|It : y | I y P P H' t t1 ~ N t|t1 , t|t1 t|t1 HP F et | It1 0 t|t1 t|t1 • Recall a property of the multivariate normal distribution:
1 1|2 1 12 22 ( x 2 2 ) 1 1|2 11 12 22 21 Then, from the joint, we can easily derive the KF update: T -1 yt|t = yt|t-1 + Pt|t-1 H (Ft|t-1) et|t-1 T -1 Pt|t = Pt|t-1 – Pt|t-1 H (Ft|t-1) HPt|t-1 29
15