Scribd
Upload
134 views3 pages

ps3 PDF

This document provides instructions and problems for an econometrics problem set. It covers topics like the linear regression model, least squares estimators, variance of estimators, partitioning sums of squares, and panel data models. For several problems, students are asked to show formulae, explain assumptions, or discuss whether regressors meet exogeneity conditions for different model specifications. The overall document provides guidance and questions to help students prepare for a lecture on the linear regression model.

Uploaded by

antonio
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
134 views3 pages

ps3 PDF

This document provides instructions and problems for an econometrics problem set. It covers topics like the linear regression model, least squares estimators, variance of estimators, partitioning sums of squares, and panel data models. For several problems, students are asked to show formulae, explain assumptions, or discuss whether regressors meet exogeneity conditions for different model specifications. The overall document provides guidance and questions to help students prepare for a lecture on the linear regression model.

Uploaded by

antonio
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

London School of Economics Vassilis Hajivassiliou

Department of Economics Michaelmas Term 2018

MSc Econometrics (Ec402)


2018–2019
Problem Set #3
The Linear Regression Model

Instructions: Prepare for week 4.

1. Consider the following equations:


(a) yt = β1 + β2 xt + t (b) yt = β̂1 + β̂2 xt + ˆt (c) ŷt = β̂1 + β̂2 xt
(d) yt = β̂1 + β̂2 xt + t (e) yt = β1 + β2 xt (f) yt = β1 + β2 xt + ˆt
 
(g) t t = 0 (h) t xt t = 0

Explain why equations (a)–(c) are correct, but (d)–(h) are incorrect. Can you correct
(g) and (h)? A diagram would be useful in answering this question.

2. Consider the simple linear regression model: yt = β1 + β2 xt + t , t = 1, · · · , T .

(a) Obtain formulae for β̂1 and β̂2 , the least squares estimators, from the general
result: β̂ = (X  X)−1 X  y.
(b) Obtain formulae for var(β̂1 ), var(β̂2 ), and covar(β̂1 , β̂2 ) from the general OLS
result: V (β̂) = σ2 (X  X)−1 .

3. Consider the simple regression model

yt = β1 + β2 xt + t , t = 1, · · · , T.

Under the standard assumptions, show that:



(a) var(β̂1 ) = σ 2 x̄2 / t (xt − x̄)2 + σ 2 /T

(b) cov(β̂1 , β̂2 ) = −σ 2 x̄/ t (xt − x̄)2

(c) E ˆ2t
t = (T − 2)σ2 .

4. The regression model yt = β1 + β2 x2t + · · · + βk xkt + t , t = 1, · · · , T, is estimated


by ordinary least squares. Define ŷt ≡ β̂1 + β̂2 x2t + · · · + β̂k xkt and ˆt ≡ yt − ŷt . Use
the normal equations to show that the dependent variable sum of squares (TSS) can
be partitioned into an explained sum of squares (ESS) and a residual sum of squares
(RSS):   
(yt − ȳ)2 = (ŷt − ȳ)2 + ˆ2t .
t t t
RSS
Hence show that R2 ≡ 1 − T SS
equals the sample correlation between yt and ŷt .

1
5. Consider the multiple linear regression model y = Xβ +  with k explanatory variables
in X. Show that if all the observations on a particular explanatory variable are multi-
plied by λ, then the residuals of the regression are unchanged while the corresponding
regression coefficient is multiplied by 1/λ. Use this result to explain what will happen
when a particular explanatory variable is measured in thousands of pounds instead of
millions of pounds.

6. Consider the classic linear regression model with dependent S ×1 vector y and regressor
S × k matrix X fitted to a panel data set, with each individual i observed over T time
periods. There are N individual in the sample, so S = N · T . The data is organized
by grouping together all time observations for each individual, so the dependent vector
is: ⎛ ⎞
y1 ⎛ ⎞
⎜ . ⎟ y11
⎜ . ⎟ ⎜ . ⎟
⎜ . ⎟ ⎜ .. ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. ⎟ ⎜ ⎟
⎜ . ⎟ ⎜ y ⎟
⎜ ⎟ ⎜ 1T ⎟
⎜ .. ⎟ ⎜ . ⎟
⎜ . ⎟ ⎜ .. ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ y ⎟
⎜ ys ⎟ ⎜ it ⎟
y=⎜ ⎟
⎜ .. ⎟ = ⎜ . ⎟
⎜ . ⎟ ⎜ ⎟
⎜ ⎟ ⎜ .. ⎟
⎜ .. ⎟ ⎜ ⎟
⎜ . ⎟ ⎜ yN 1 ⎟
⎜ ⎟ ⎜ ⎟
⎜ . ⎟ ⎜ ⎟
⎜ .. ⎟ ⎜ y N 2 ⎟
⎜ ⎟ ⎜ . ⎟
⎜ . ⎟ ⎜ ⎟
⎜ . ⎟ ⎝ .. ⎠
⎝ . ⎠
yN T
yS
The regressor matrix X and the true error vector true are organized in a similar fashion.
The model is assumed to satisfy:
A1 : rank(X) = k
A2 : y = Xβ true + true with Etrue = 0
The regressors X are assumed to be unrelated to the error term true in one of two
alternative ways:
Strongly exogenous defined as: A3Rmi : E(true |X) = Etrue
W eakly exogenous defined as: A3Rsru : E(true
s xsj ) = 0 for every regressor variable
j.
Finally, it is believed that
(i) errors true true
it and ns are independent for i = n and any t, s, and

(ii) errors true true


it and is are correlated for all i and any t, s.
Consider the following three cases for model specification. In each case, discuss
whether the regressors will be Strongly Exogenous with respect to the error term,
and whether or not they will be Weakly Exogenous:

2
(a) The model describes an experimental situation so all regressors are non-stochastic,
i.e., fixed in repeated samples.
(b) The jth regressor is the first lag of the dependent variable, i.e., xjit = yi,t−1
(c) Each regressor for individual i in period t is independent of the corresponding
error for that individual i in the same period t, i.e., with true it , but it may be
correlated with the past values of that individual’s error, i.e., it may be correlated
with true
is , s < t.


c Vassilis Hajivassiliou, LSE 2000-2018

You might also like