Вы находитесь на странице: 1из 2

additional problem 2

d
Let X1 = Y1 − Y3 , X2 = 2Y1 + Y2 − 2Y3 and X3 = −2Y1 + 3Y3 , where Y1 , Y2 , Y3 = NIID(0, 1).
d
Show that X2 |(X1 + X3 = x) = N(−6x/5, 9/5).
Solution. Since X1 + X3 = −Y1 + 2Y3 ,
 
 Y
2 1 −2  1 
    
X2 2Y1 + Y2 − 2Y3
= = Y2 .
X1 + X3 −Y1 + 2Y3 −1 0 2
Y3
 
d Y1 d
Since Y1 , Y2 , Y3 = NIID(0, 1), they are jointly normal with Y2 = N(03×1 , I3×3 ). Hence, by the
Y3
reproducing property of gaussian random vectors under linear transformations,
      0
X2 d 0 2 1 −2 2 1 −2
= N( , ).
X1 + X3 0 −1 0 2 −1 0 2
Now,  
  0   2 −1  
2 1 −2 2 1 −2 2 1 −2  9 −6
= 1 0 =
 .
−1 0 2 −1 0 2 −1 0 2 −6 5
−2 2
Hence,      
X2 d 0 9 −6
= N( , ).
X1 + X3 0 −6 5
Since X2 and X1 + X3 are jointly normal,
d
X2 |(X1 + X3 ) = N(BLP[X2 |X1 + X3 ], var(X2 − BLP[X2 |X1 + X3 ])).
Now, BLP[X2 |X1 + X3 ] = β0 + β1 (X1 + X3 ) with β0 = EX2 − β1 E(X1 + X3 ) = 0 and
cov(X1 + X3 , X2 ) cov(X1 , X2 ) + cov(X3 , X2 ) 6
β1 = = =− ,
var(X1 + X3 ) var X1 + var X3 + 2 cov(X1 , X3 ) 5
because cov(X1 , X2 ) = cov(Y1 − Y3 , 2Y1 + Y2 − 2Y3 ) = 4, cov(X3 , X2 ) = cov(−2Y1 + 3Y3 , 2Y1 + Y2 −
2Y3 ) = −10, var X1 = 2, var X3 = 13, cov(X1 , X3 ) = cov(Y1 − Y3 , −2Y1 + 3Y3 ) = −5. Hence,
BLP[X2 |X1 + X3 ] = −6(X1 + X3 )/5, which implies that
6
var(X2 − BLP[X2 |X1 + X3 ]) = var(X2 + (X1 + X3 ))
5
36 12
= var X2 + var(X1 + X3 ) + cov(X2 , X1 + X3 )
25 5
36 12 9
= 9 + 5 + (−6) = .
25 5 5
Therefore,
d 6 9 d 6 9
X2 |(X1 + X3 ) = N(− (X1 + X3 ), ) =⇒ X2 |(X1 + X3 = x) = N(− x, ).
5 5 5 5

1
2

additional problem 3
Let X and Y be random variables such that
     
X d 3 1 −1
=N , .
Y 1 −1 4
(i) Define what it means for X and Y to be jointly normal.
(ii) Find the joint distribution of X + 2Y and 2X − Y .
(iii) Calculate Pr(X + 2Y ≤ 2X − Y ).
(iv) Find E[Y |X] and E[Y |Φ(X)].
Solution. (i) X and Y are jointly normal if every linear combination of X and Y is a Gaussian
random variable.
(ii) Let Z1 := X + 2Y and Z2 := 2X − Y , and note that
      
Z1 X + 2Y 1 2 X
= = .
Z2 2X − Y 2 −1 Y
Hence, by the reproducing property of Gaussian random vectors under linear transformations,
        0     
Z1 d 1 2 3 1 2 1 −1 1 2 d 5 13 −9
=N , =N , .
Z2 2 −1 1 2 −1 −1 4 2 −1 5 −9 12
(iii) Using the definitions of Z1 and Z2 in the previous part, note that
Pr(X + 2Y ≤ 2X − Y ) = Pr(Z1 ≤ Z2 ) = Pr(Z1 − Z2 ≤ 0).
d
Since (ii) reveals that Z1 and Z2 are jointly normal, it follows that Z1 − Z2 = N(0, 43). Hence,
Pr(Z1 − Z2 ≤ 0) = Pr(N(0, 43) ≤ 0) = 1/2.
Alternatively, note that Pr(X + 2Y ≤ 2X − Y ) = Pr(3Y − X ≤ 0). Since X, Y are jointly normal,
 
  Y d
3Y − X = 3 −1 = N(0, 43).
X
Hence, Pr(X + 2Y ≤ 2X − Y ) = Pr(3Y − X ≤ 0) = N(0, 43) = 1/2.
(iv) Since X and Y are jointly normal,
d
Y |X = N(BLP[Y |X], var(Y − BLP(Y |X))),
i.e., E[Y |X] = BLP[Y |X] and var[Y |X] = var(Y − BLP[Y |X]). Now, BLP[Y |X] = β0 + β1 X, where
β1 = cov(X, Y )/ var(X) = −1/1 = −1 and β0 = EY −β1 EX = 1−(−1)3 = 4. Thus, E[Y |X] = 4−X.
Moreover, since Φ is injective on R, E[Y |Φ(X)] = E[Y |X] = 4 − X.

Вам также может понравиться