7/7/24, 2:57 AM Linear Regression Properties

Let's address the questions in detail.

4. Consider the model

yt = αxt + ut ; t = 1, 2, … , T
​ ​ ​

where E(ut ) ​ = 0, E(u2t ) = σ 2 xt , E(us ut ) = 0 if s =

​

 t, for all s and t. The density function for ut is ​ ​ ​

2
f (ut ) = ​ (2πσ 2 x2t )−1/2 ​ exp [− 2σ1 2 ​
( uxtt ​
​

​
) ].

Derive the maximum likelihood estimators of α and σ 2 .

Maximum Likelihood Estimation:

The likelihood function for the sample is:

2
T T
L(α, σ ) = 2
∏t=1 ​ f (ut ) = ​ ∏t=1 (2πσ 2 x2t )−1/2 ​ ​ exp [− 2σ1 2 ​ ( uxtt ​
​

​ ) ].

Substituting ut ​
= yt − αxt , we get:​ ​

2
T yt −αxt
L(α, σ 2 ) = ∏t=1 (2πσ 2 x2t )−1/2 exp [− 2σ1 2 ( ​ ​ ​

xt
​

​
​

​ ) ].

Taking the natural logarithm to get the log-likelihood function:

2
ln L(α, σ 2 ) = − T2 ln(2π) − ​
T
2
​ ln(σ 2 ) − 12 ∑Tt=1 ln(x2t ) − ​ ​ ​
1
2σ 2
​ ∑Tt=1 ( ​
yt −αxt
xt
​

​
​

​ ) .

Simplify the log-likelihood function:

T T (yt −αxt )2
ln L(α, σ 2 ) = − T2 ln(2π) − T
ln(σ 2 ) − ∑t=1 ln(xt ) − 1
∑t=1 .
​ ​

2 2σ 2 x2t
​ ​ ​ ​ ​ ​ ​

To find the maximum likelihood estimators, we take the partial derivatives with respect to α and σ 2
and set them to zero.

Partial derivative with respect to α:

∂ ln L T (yt −αxt )(−xt ) T yt −αxt

= − σ12 ∑t=1 = 1
∑t=1 .
​ ​ ​
​ ​

∂α x2t σ2 xt
​ ​ ​ ​ ​ ​ ​

​
​

Setting the partial derivative to zero:

T yt −αxt
∑t=1 ​
​

xt ​
​

​ = 0.

https://chatgpt.com/c/0e332593-feba-456f-a623-4a5a45bb8bf4 1/3
7/7/24, 2:57 AM Linear Regression Properties
T T
∑t=1 yxtt ​
​

​
​ = α ∑t=1 ​ 1.
yt
∑Tt=1
​

α= xt
.
​ ​

T
​

Partial derivative with respect to σ 2 :

∂ ln L T (yt −αxt )2
= − 2σT 2 + 1
∑t=1 .
​ ​

∂σ 2 2σ 4 x2t
​ ​ ​ ​ ​

Setting the partial derivative to zero:

(yt −αxt )2
− 2σT 2 + ​
1
2σ 4
∑Tt=1 ​

x2t
​

​
​

​ = 0.

T 1 T (yt −αxt )2
2σ 2
​ = 2σ 4
​ ∑t=1 ​

x2t
​

​
​

​
.

1 T (yt −αxt )2
σ2 = ∑t=1 .
​ ​

T x2t
​ ​ ​

Thus, the maximum likelihood estimators are:

y
∑Tt=1 xt
​

α^ = T t ,
​ ​

T (y −α^x )2
σ^ 2 = T1 ∑t=1 t x2 t .
​ ​

​ ​ ​

t ​

5. Consider the following data generating process

Yt = ϕ1 Yt−1 + ϕ2 Yt−2 + εt + θ1 εt−1 + θ2 εt−2 + θ3 εt−3
​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

where εt is a white noise process. Assuming −1​ < θ2 < 0 and without taking a first difference, answer
​

the following:

(a) Find the h-step-ahead forecast for Y for h = 1, 2, … , k .

The h-step-ahead forecast is given by:

Y^t+h∣t = E[Yt+h ∣Ωt ] ​ ​ ​

Where Ωt is the information set at time t. ​

For h = 1:

Y^t+1∣t = ϕ1 Yt + ϕ2 Yt−1 + θ1 εt + θ2 εt−1 + θ3 εt−2

​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

For h = 2:

https://chatgpt.com/c/0e332593-feba-456f-a623-4a5a45bb8bf4 2/3
7/7/24, 2:57 AM Linear Regression Properties

Y^t+2∣t = ϕ1 Y^t+1∣t + ϕ2 Yt + θ1 ε^t+1∣t + θ2 εt + θ3 εt−1

​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

Since ε^t+1∣t ​ = 0:

Y^t+2∣t = ϕ1 (ϕ1 Yt + ϕ2 Yt−1 + θ1 εt + θ2 εt−1 + θ3 εt−2 ) + ϕ2 Yt + θ2 εt + θ3 εt−1

​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

For h = k:

Y^t+k∣t = ϕ1 Y^t+k−1∣t + ϕ2 Y^t+k−2∣t + θ1 ε^t+k−1∣t + θ2 ε^t+k−2∣t + θ3 ε^t+k−3∣t

​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

(b) Find the h-step-ahead forecast error for Y for h = 1, 2, … , k .

The forecast error for h = 1:

et+1 = Yt+1 − Y^t+1∣t = εt+1

​ ​ ​ ​

The forecast error for h = 2:

et+2 = Yt+2 − Y^t+2∣t = εt+2 + ϕ1 εt+1

​ ​ ​ ​ ​ ​

The forecast error for h = k:

et+k = Yt+k − Y^t+k∣t

​ ​ ​

(c) Find the h-step-ahead forecast error variance for Y for h = 1, 2, … , k .

The forecast error variance for h = 1:

Var(et+1 ) = Var(εt+1 ) = σ 2
​ ​

The forecast error variance for h = 2:

Var(et+2 ) = Var(εt+2 + ϕ1 εt+1 ) = σ 2 (1 + ϕ21 )

​ ​ ​ ​ ​

The forecast error variance for h = k:

2(k−1)
Var(et+k ) = σ 2 (1 + ϕ21 + ϕ41 + … + ϕ1
​ ​ ​ )
​

For a more general form, the variance would include terms related to all the ϕ and θ parameters,
capturing the structure of the process up to h steps ahead.

https://chatgpt.com/c/0e332593-feba-456f-a623-4a5a45bb8bf4 3/3