7/7/24, 2:54 AM Linear Regression Properties

Let's address the questions in detail.

1. Consider the single equation model

yi = z i β + ε i
​ ​ ​

 0. Let β^ denote the IV estimator of β using

where yi and zi are both real-valued (1 × 1) and E[εi ∣zi ] =
​ ​ ​ ​

as an instrument a dummy variable di (takes only the values 0 and 1). Derive the explicit expression for ​

the IV estimator.

IV Estimator:

The IV estimator for β is given by:

β^IV =
∑ i d i yi
​ ​ ​

∑ i di z i
​ ​ ​

​ ​ ​

This uses di as an instrument for zi . ​ ​

2. Consider the following data generating process

Yt = εt + θ2 εt−2 + θ3 εt−3
​ ​ ​ ​ ​ ​

where εt is a white noise process with V (εt )

​ ​ = σ 2 . Assuming −1 < θ2 < 0 and −1 < θ3 < 0, answer
​ ​

the following:

(a) Derive the expected value of this process.

The expected value of Yt : ​

E[Yt ] = E[εt + θ2 εt−2 + θ3 εt−3 ]

​ ​ ​ ​ ​ ​

Given that εt is white noise with E[εt ] ​ ​ = 0:

E[Yt ] = 0 ​

(b) Derive the variance of this process.

The variance of Yt : ​

Var(Yt ) = Var(εt + θ2 εt−2 + θ3 εt−3 )

​ ​ ​ ​ ​

Given that εt is white noise with variance σ 2 and εt are uncorrelated for different t:
​ ​

Var(Yt ) = σ 2 + θ22 σ 2 + θ32 σ 2

​ ​ ​

Var(Yt ) = σ 2 (1 + θ22 + θ32 )

​ ​ ​

https://chatgpt.com/c/0e332593-feba-456f-a623-4a5a45bb8bf4 1/5
7/7/24, 2:54 AM Linear Regression Properties

(c) Derive the covariance of this process for all j . Note that j represents the number of periods
between Yt and Yt−j . ​

The covariance between Yt and Yt−j : ​ ​

Cov(Yt , Yt−j ) = E[(Yt − E[Yt ])(Yt−j − E[Yt−j ])]

​ ​ ​ ​ ​ ​

Given that E[Yt ] ​ = 0:

Cov(Yt , Yt−j ) = E[Yt Yt−j ]

​ ​ ​

For j = 1:

Yt = εt + θ2 εt−2 + θ3 εt−3
​ ​ ​ ​ ​ ​

Yt−1 = εt−1 + θ2 εt−3 + θ3 εt−4

​ ​ ​ ​ ​ ​

Cov(Yt , Yt−1 ) = E[(εt + θ2 εt−2 + θ3 εt−3 )(εt−1 + θ2 εt−3 + θ3 εt−4 )]

​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

Since εt are white noise and uncorrelated for different t:

​

Cov(Yt , Yt−1 ) = 0 ​ ​

For j = 2:

Yt = εt + θ2 εt−2 + θ3 εt−3
​ ​ ​ ​ ​ ​

Yt−2 = εt−2 + θ2 εt−4 + θ3 εt−5

​ ​ ​ ​ ​ ​

Cov(Yt , Yt−2 ) = E[(εt + θ2 εt−2 + θ3 εt−3 )(εt−2 + θ2 εt−4 + θ3 εt−5 )]

​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

Cov(Yt , Yt−2 ) = θ2 σ 2
​ ​ ​

For j = 3:

Yt = εt + θ2 εt−2 + θ3 εt−3
​ ​ ​ ​ ​ ​

Yt−3 = εt−3 + θ2 εt−5 + θ3 εt−6

​ ​ ​ ​ ​ ​

Cov(Yt , Yt−3 ) = E[(εt + θ2 εt−2 + θ3 εt−3 )(εt−3 + θ2 εt−5 + θ3 εt−6 )]

​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

Cov(Yt , Yt−3 ) = θ3 σ 2
​ ​ ​

For j  2 and j =
=  3:

Cov(Yt , Yt−j ) = 0 ​ ​

(d) Derive the autocorrelation function for all j .

The autocorrelation function ρ(j):

https://chatgpt.com/c/0e332593-feba-456f-a623-4a5a45bb8bf4 2/5
7/7/24, 2:54 AM Linear Regression Properties
Cov(Yt ,Yt−j )
ρ(j) =
​ ​

Var(Yt )
​

For j = 2:
θ2 σ 2 θ2
ρ(2) = ​

σ 2 (1+θ22 +θ32 ) ​
​ = 1+θ22 +θ32 ​
​

​
​

For j = 3:
θ3 σ 2 θ3
ρ(3) = ​

σ (1+θ22 +θ32 )
2 ​
​ = 1+θ22 +θ32 ​
​

​
​

For j  2 and j =
=  3:

ρ(j) = 0

(e) Plot the autocorrelation function with the information you derived above.
To plot this:

1. Draw the x-axis representing the lag j .

2. Draw the y-axis representing the ACF values ρ(j).

Points:

At lag 0: ρ(0) =1
At lag 1: ρ(1) =0
θ2
At lag 2: ρ(2) = 1+θ22 +θ32 ​ ​
​

θ3
At lag 3: ρ(3) = 1+θ22 +θ32 ​ ​
​

For higher lags: ρ(j) =0

3. Consider a product market with a supply function Qsi ​ = β0 + β1 Pi + usi , a

​ ​ ​ ​

demand function Qdi = γ0 + γ1 Pi + udi , and a market equilibrium condition ​ ​ ​ ​ ​

Qsi = Qdi , where usi and udi are mutually independent i.i.d. random variables,
​ ​ ​ ​

both with a mean of zero and variance σs2 and σd2 , respectively. ​ ​

(a) Show that Pi and usi are correlated. ​ ​

From the market equilibrium condition Qsi ​ = Qdi :

​

β0 + β1 Pi + usi = γ0 + γ1 Pi + udi
​ ​ ​ ​ ​ ​ ​ ​

Solving for Pi : ​

https://chatgpt.com/c/0e332593-feba-456f-a623-4a5a45bb8bf4 3/5
7/7/24, 2:54 AM Linear Regression Properties

(β1 − γ1 )Pi = γ0 − β0 + udi − usi

​ ​ ​ ​ ​ ​ ​

ud −us
Pi = γβ01−−βγ01 + βi1 −γ1i
​
​ ​

​ ​ ​

​ ​ ​ ​

The price Pi is a function of both usi and udi . Given that usi appears in the denominator, Pi and usi are
​ ​ ​ ​ ​ ​

correlated.

(b) Show that the OLS estimator of β1 is inconsistent. ​

Inconsistency arises because Pi is correlated with usi . In the supply equation: ​ ​

Qsi = β0 + β1 Pi + usi
​ ​ ​ ​ ​

Since Pi is correlated with usi , the OLS estimator of β1 will be biased and inconsistent because one of
​ ​ ​

the OLS assumptions (no correlation between the regressors and the error term) is violated.

(c) How would you estimate β0 , β1 , and γ0 ? ​ ​ ​

To estimate β0 , β1 , and γ0 , we can use an instrumental variables (IV) approach. A possible instrument
​ ​

could be a variable that affects supply but not demand, or vice versa.

For example, if we have an instrument Zi that affects supply but not demand, we can use two-stage ​

least squares (2SLS):

1. First stage:

Regress Pi on the instrument Zi : ​ ​

P i = π 0 + π 1 Z i + vi
​ ​ ​ ​ ​

2. Second stage:

Use the predicted values P^i from the first stage in the supply equation: ​

Qsi = β0 + β1 P^i + usi

​ ​ ​ ​ ​

For the demand function:

1. First stage:

Regress Pi on the instrument Zi (if another instrument for demand is available):

​ ​

P i = π 0 + π 1 Z i + vi
​ ​ ​ ​ ​

2. Second stage:

Use the predicted values P^i from the first stage in the demand equation: ​

https://chatgpt.com/c/0e332593-feba-456f-a623-4a5a45bb8bf4 4/5
7/7/24, 2:54 AM Linear Regression Properties

Qdi = γ0 + γ1 P^i + udi

​ ​ ​ ​ ​

By using instruments that are correlated with Pi but uncorrelated with the error terms usi and udi , we
​ ​ ​

can obtain consistent estimates of β0 , β1 , and γ0 .

​ ​ ​

https://chatgpt.com/c/0e332593-feba-456f-a623-4a5a45bb8bf4 5/5