Caleb Ziblim

Econometrics I
2023
University of Ghana
 Introduction
Pi + B2Ki stochastic
dependent
> no
varichte
y y
-
=
-

variable B + Periter Stochastic

independent
> -

2 =

y
-

Simple regression model

⑨: +
B +
o
wa = 0

abnat interp.
Variest
Shope
Coffcut

using
-
OLS
;
Bo E t
y
=
-

51 =

Enigi =
& (x -

m) Ly j)
-

En ; 2 2(n -

i) 2

Example
N

3
Y
in y
-

5
Engi 4
↑
2
-

5 O 0 O
2

7
6 2
j 4 8

E(n -) =
0
Elz-5) =
0 Enigi 4
(n-rs2 = o =

i 5 6
y
= =

Zi =
B1 = 0

Yo B n 6
j 0 5(5)
: -

,
=
- .

= 3 .
5
-
-M

: Bo + P n

y
=
,

3 5 + 0 5m
y
=
.
.

TSS RSS Ely-y)2

2
=
z(y j) -

= = Seei
Ess 2

&(j j)
= -

Caleb Z
 11 and should be unrelated
always
-
in

line
- a
larger intercept (BO) Sifts the
regression upwards
intercept cuts the below the
a
negative y
atis in atis
~

when B , =
0 it is a horizontal line
S
P1 Line
large Steeper regression
-
=

Objective of regression analysis

O G establis thes relations p
if a

O forecast new value

finaling wa

-2 how well
Fells
regression line
predicte actual des
a
us
-

22 distance the actual the

-

Compares the stwon value to men

2 the value to the

estimated mean

r
G(j j
=
= =
-

z(y
-

yjz
01 r 2 /

it w = 0
-
> no re l a t i o n s
p

we = m Perget relationslip

Examquestion : If
any of
the assumptions fail
,
will we

be able to estimate te
parameters ?

Variance Estimators
of
Exam question : WhatCondition Can the variance be heteroskedstic ?

spread
it is
hint ; about the
Eg
.
a
highy unequal population

=
VarLu) = 02

~ Parameters
no .
Of
 -Ge
var (B ) .
=

Ep

Excu
question
.
When can the variance
of , be
large ?
Hint when variance is
; the e r ro r
larg
-
ie
. Var(B , ) = G2
-

2 (ni =

n)2

Sum residuals must be o

of

add
Why do we the e r ro r term ?

0 Omittel variables

① factor onlinearities
⑦ Measurement Errors

⑦ Unpredictable effects

Linearity In
Simple regression
9 Must be knear in variable

y=
B + Bex
, + e - near bes 23/2n = 32
B Pante > ha Sos My/on 2
Pa
y
-

=
.
+ =

= B + Be "n + e - c m kd dos Islan Ba/m "

y
-

,
=
,

Unear the is constant

means
stope
With Linear variables can knear but
regression /
be n on pracemeters
must always be linear .

Linearity
.
3
in Parameters
=
B, + B2x +
- >
Liner in
Parameters
y
u

y =
B ,
+ Pzm + m -
>
nonlinear in Parameters
Properties of regression
line

0 It through he in
i
sample means <
passes of
=F
-

O mean
value
of
actual ↑ d estimated Y a re
equal ie
③
=
residuals
Zeti
Mean va l e is 0 ie 0 0
of =

Assumptions of Classic Linear

Regression Model

O Model linear
parameters
in
is

& Thre has to be

variability in the Sample (n)
regressor
⑤ The va r i a n c e
of each mi is constant/homoscedetic
# There is no autocorrelation ston 2 a r r ow terms it. random

⑤ The model specified

regression
is
crrety Specifiction bies
.
i e .
no

including
-
not an
including an

important
i r r e ve r t

variabi varidie

I
~
makes the ols
Estimator biasel !!?

(source has
of
bas
depends on the
relationsp
the ommtil varidle with u

Unbiasedness os
of estimators

Y = B + Pen + u
,
-

:
2(x - i)
(y -

5) = E(x - m) [CP .
+ Pan + uS -

(1 ,
+ Ban +
in)]
San-a)
=
[Pala-n) + Leu-e)]
= BrELn-5] +
[In-5) (ee-tes
M

B2 = =
Be E (ni -2) + &(ai -

i) (ei -

i)
2 (oi -)2
 Be
> 0
= B2 + E(mi - ) ( e e)
-
Since Eccl = o

E(mi -
n jm

Da =
Beautier

Gauss-Markov Theorem

States that the Ols estimator is BLUE (Best Unear Unbiase Estimated

if If
the holds
;
a It is a lineer function (in Parameters)
6) It is unbiased in Elise) = Be

6) It has a minimum variance in the class

of
allSuch Linea

nubiava Estimators -

.
Note Var (eci) = 02

==
wohre or

,
Ver (Bel
=
this implies Not
;

O the the variation in the smaller the Bed vice

larger
w
use
variance
,
of
O the ord the
larger the ,
larger the va r i a n c e
of
Be

The logn
the
Sample Size
,
he mor
varility in i a the
Bm is
m o re accurate

Caleb Z
 iss = Ess + RSS

Ely -j(z z(j y(2 Ely - y)2

=
-
+

R=
1 =
#has to be variation in m

02
= =
Varie) =

Var(s)
=
↓
sections the
er e r ro r variance (2) the
larger he va r i e
a
5
① The (El-5(2)
larger the
variability in ni the Smaller te

variance
of
Bi

Se()
==
Multipla near
regression
Bo + B, + Bene Blak + e
y = ms . .

The independent variables not be Lie Collinearity)

-

must related .
.
multi

Matrix to
finding
Gegficients
approach
P = (n'n) +
my
 +
var/cuB = 0 (n'n)

[ corpais]
I Var B CovBiBe CovB , By

Lov Be B , Var Be

GovB3 Bi CovPbP2 Var B3

of =
where it

Example

·3
5

g
I 9

10 I+men te

B = In')Y
-
-
-

= ↓
3 12 x1 I I

Y
2 I I

12 8 9
Y
I 8

I 7 14 6 8
-

5 I 9 7

g g
-- -
I L
-

I'ml"
i win, adj
= + (22)

[2 [ &
(n'n) (m'2)"

I
5 46 485

=5
0 956 0

a
- .
- .

46 0 .
088 0-029

166 0 485
-

0 029 0 042
.

-
-

Caleb Z
 [ii]
is =

154

[s405 &
P (2)
my
=
0 956
-

=
.

-0 .

Soling Maticus with Gluctor

Steps :

↓ Mode -
> Matrix-1 Wat A

② selet dimension first matric (3 + 3)

of
③ Press and
Input numbers and
Shift 4
① Select Data
⑤ Mat is and
repeat 2
⑧ Press AC

②Shift & 4
⑧ Select Mat A add
operation
a select Mat B

& Press 1
=
1

iI(j) 2
=

: ] [s 120 ] [i]
= 24 +
=
+ 1s

(2 + 1) + (1 + 4) + (2 + 3) 2 + 4 + 6

[ J i
Using Carc A B =
;

12
 R2
1-
R" =
- -
Adjusta
= 1 =

Broscrea in M ESS
Things to ease
note
; -
R2 will increase when another
independent variable is added
-

This makes R22 unreliable

-

If added the
parameter estimates
unnecessary variables a re
,

remain Same
he

If
importat variable exclued,
an is the wo u l
parameters
-

be bissel .

Hypothesis testing

Additional assumption;
The er ror term is
normally distributed with a mean
of o
variance
ot
of

O Testing whether individual

Gefficients a re
different from
/equal to
D

hy
-
State auth a alternate potes's

Ho Pi Hi :
Bi > ch i ld fort
-

:
=
P 0

We're
testing to Check
if the Coefficients are
Significantly different
from 0 . This means the
independent variable influences to

depecht varichte

Find
-

Calculates t valu

=
teal
-

Find Critical t values

↓ crit
tin-
= or terit = ta n-k
,

& - to be tet

2 tale
+ st

Decision Rule

if It call > Itcrit) -

>
Ho
reject
if It cal 1 2 Itcrit >
-

Ho
accept

if (Itall Harity) it
rejetH o the
deficits
we >
means
I
are
Significat and Tur is a linear relation sp won
mid
y

Note : P
Any %, insignificant
is
value > 10 means
Coefficient
.

Confidence Interval

x =
=SecEs-margin
nex +

of
aro-

where Seli Starti

Fra
= =

② Testing if begricet is to Constant

equt a

State alternate
hypotees
-

null d

B
a .
Ch T2
Ho : , = al H1 : B aa

Conat at

Find Glalete + values

take

=is
a
-
find Critian + values

-crit farit to
+
= n - k or = n-k
, ,

Decision Rule
-

if It call > Itcrit) -

>
Ho
reject
if It call Itcrit -
> Ho
accept

if (Italc) B ai
we
accept
Ho < Itcrizh) it meance
=

③ Testing if Coefficit is
eque
to a parameter ; (linear Combinating
State null alternate
hypothes
-

&

Ho : Bi = Pr H1 : B1 + Be

find Glalete +

=
t calc

31 -

Be

find Initial t

twit =
ty ,
n-k or to ,
nuk

Decision Rule

if It call > Itcrit) -

>
Ho
reject
if It cal 1 2 Itcrit >
-

Ho
accept

This is used to test for constant returns to sale

Caleb Z
 SelB -Bal
finding ,

var(B 1
-

B2) = Var(P1) + Var(B2) -

2 Cor (B , Ba)

SeLii-Bel JSelB
Selpe-
12)
2
= . ) +
,

④ Testing Exclusion Restrictions

g test is as & 24 have no

eyet so that we
only
know whether to include them or not

state alternate
by pothesis
-

nul a

Ho : By = By = 0 Hi : By = p

Means the 417 moded C

Old mode is in valid
is Valid

Using
f test Calculatet h e Falc
-

sch of restricted

-
feala :
-SSR)/@noo reviations

ur/n-Q
I T Of
Sa
of Parameters
Unrestrictel

-titled
Unrestriea
↓
+ meas + 4424 + u

SSR this is SScur

of

Po + B 2 + Bere +e
y
= , ,

- >

SSR Of Mis is SSRr

-

Final Critical F values

Feriz = F2 n K
g ,
-

Decision Rule

if Falc Farit
> -
>
Reject Ho

Fal < Fart >

-
Accept Ho

⑤ Testing for Overall

Significence
used to
jointly test if all the
Stope Deficients a re
el to o

State alternate
hypoless
-

null d

Ho : B1 = B2 .
. .
.
BK =
w

Bo B Panz B3x3
given ; yi
= + , x + + + e

Men
;
Ho :
B
,
= Bz = Bz = 0 It i . Bit 0

find Calacatel F

Faceb
Tin -
1)

final Critical F

Frit = Fa k -
1 n -
1
, ,

Decision Rule
-

if Falc Farit
> -
>
Reject Ho

Fal < Fart >

-
Accept Ho
Example
sonhotansthis paints (Sjnfica gitia) - T +

5) Overall I
Synigence -
> tot

PRF :
On +
Born + u

N
Taxing for tis individuly
we use i to t

>
Ho : B2 By -
tot for
Spigene
= = 0 our

Ya
Hi Br B + 0

Ho

H1
:

:
inctively
R > o
-> Tony for
goodness of
fo

for
Testing individua
Significance of
B1 & Be does not mean that the

RF
overall is
Significant
· for individual 2 F for
we need to Conduct T tot
Signifiance Not

Import Previous
Overall
Synigiance GDP am Prices Import

d
accrry
-
Price

age
Se =
=

10
1

.
.
073

797)
+ 5 .

(0
20 .
812)
/ 20 .

111
+ 0 .
054me

10 22)
.
+ 0 .
045M2
20
G
.

019)
R2 = 0 .

934 als

t = 30

⑦ Interprete the results

Interprete all the

Geggaints & R2
0 1 .
073

When all the variables the Per

independent a re o
, anyage wage
entlogee is 1 .

073 units

① 5 .
288

VE =
unfilled Job
vacanc
: If
unfilled Job by
Vanary increases a unit
,
the the
Aug
Wages will increase
by 5 . 288
holdy or
independent various

Justar t

Same for other variates

indepecht
-

interpretation
.

22 =
0 .

934
Means that 93 4 % totch variation in
dependent vari at i on is
of
.

effend by the
independent variation in the model

⑤ Which the
Significat
which
begicets
estimatel and
a re
o
of
the various a re
dropped .

Note that
If deggicist is
insigigiant , cop variable
.

any

+ tex ; Be
Using Since we
k n ow is Patie

for

o
Be

Ho : B2 = 0 H .: Ba & O

Calculated
- =

E-0
=

Critic + = +c n-k = to .
os 30 - 5 =
to os 25 = 1 .
708
, , ,

Clearly ,
Calculated t > Critic t

: We Ho
reject
and hence Bo
,
is
statistically Sigrigct

Caleb Z
② What is the rationale
of using import
Prices in the mode !

Mt < M+ a re
import price
-1

Because , increase prices with Go

living which
import
in increase of

world
put pressure on
wages
to increase . Hace there is a
positive
relationship betwn wases >
import price
-

2) Tea for overall

signifiance the model .

of
we need R2 < N(t)

Sochi
-

R2 = 0 .

934 k = 5

n = 30

Ho
-
·

R = 0 Hi : r2 > o

=
dini /4
=
clacate = = 8

Critian F = F005(K-1) (n-k) = 7005(4 25) =

, ,

Glaldel F > Critian F

: We
rejet
the mor
hypotes's
fit
Hence
,
model is
Statistically Significant and thes a
good

Incremental/Marginal Contribution
of
an
explanatory variable
T is to m e a s u re n ow diffect the model is when a new

model)
independente variable is added (ie it becomes a re a c t e d

We use He F test here

,
sch of restricted

-
faala =
-SSR)/@noo reviations

ur/n-Q
I &
roof
SSR
Of Parameters
Unrestrictel
 or

We Can use is" When it is

given

Falante = (Reno-RE)/g
41 -)/(n
&
- R u
-

k)

Same F Critical =
Fr(q ,
n - )

bi
a R2 de not makeh e mode m o re rich . Boss

models with m o re indevout Variables the RP is bond to be higher

,

"

:. in
We can non
adjusted
R2
As s
= 1 -

1-
Adjustd RE (2) =

R2
1 /
- =ll-
: =
or

"
i can be
negative
-

Note
:

for
a 2 va r i c u e
regression
modul
, testing the

inclivich
Significance of Gefficients is
enough
aral tot for
Ovec'synificance is not relevant
 Testing equality of
2
regression Geggicients (Convent returns)

P + Pede + Bynx + Byx4 +

y
= . ei

Ho
were
tating if
B3
:
By =
B4 or -

B4 = 0

I
It 13
+ By Or 33 B + = 0
-

ie. if
Were
checking the
Greggicients of 23 & My a re
equal
+ test
Using ;

En
Calculated + test -

Calculated + = <B - 4)
-

Se < 13 -

By)

: +
a 4)

critical t =

t ,
n +

is4)
firling Se
J
n

Selig - ) = arciss P4) -

= ↓ Variss + var By-2 Cris Esy)

,

M
Example
141 76 + 63 472 12 962
2
9343
y
= . .

.
- .
+ 0 .

(6 37)
.

(4 .
77) 20 98) .
10 .

05)
Gr(3 4) ,
= 0 .

0576
R2 =
0 .

9983 n = w

TestIf e

need ofaddintheLudictermt
there is
any
is Es = By · is is
=> Ho : By- By =
0

4. : By -
By 0

io- 34
+=
Clatel

4) Swanse + varix -

20 15
,
84)
= ~
12 .

96 -
0 -

93 =
-

13 .

3/

J(0 .

902 + 0 .

052) -
- 0 .

0576

A critical n

t
=
,
 Caleb Z

Dummy Variables

The a re used to test the

impact of qualitative variables or

Variables
quantitative
.

>
dependent
.

variable
quanitative
-

i
.
e

Independent- qualitative variable

dummy varidre takes the value

of 0 20
S &
Other wher observation has
wise
the devic character

Eg .
male 2 female

E
Di = & = Mate

0 = otherwise (females

Education lauch :

No education & = me ice]

We
only inclu Q- I dummies in the model
regression
The Omitted Category / is the intercept
number
of Categories
ie.
One
Category is left out .

The left
Category out is the
reference/base ategory
The base be
Category can
category
-

any
.

Exampleeffect
-

of gender on
Earnings
Bo B
Earnings = + , Gender + B2
Experience
but because Gender is
qualitative ,
we use
dummy variables

-Gender has female

categories Make
2 : .
e a

un-1
So
dummy
I
we use
-

We can either
gender
-

use
 Bo
Earnings B2E +
P,
: = + Male 2) +

where

=
I
-

I I
? =
i I
n

y
=

33
10 I

for Parameters
after soling
12
;

Earrings of
a female : Po +
Pexp
Earrings Po P1 P2
+
a male : + Exp
of

Interpretation of
dummies
;
B is sturn female
difference
the the male and
earnings
-

, a
of
a

with the same led

of experience
base when
Always make to the
category interpreting
-

reference
Alternative Interpretation : Maly the lbor market ecorm
-

in

B (orless than females in the Sama market with

negativel
, m o re
g

the same back

experience
-

Always ferform Significant a tet on the

Gefficients first
.

Generating b dummies

Eduction Level : No eduction - > D ,

Pretertiary -
> D

Tertiary 13
-
>

butSince In-1) clumin have to

can
only use
we we
,

exclude one
Category [ no eduction is our basa
group here)
- Our
Regression moder
;
By
Earnings Bo + P
Pretertiary Batertiary Exp +
+ + e
= ,

Pretenting & trticy-1 G

 petaithe difference stun the
earingings Stwn Geome with

education (bace
Pretertiary education and someone with no

group

: B2 is the
difference stun he
earnings of
someone with

election
tertiary education I someone with no (bace
groups

Note O u r :

results won't be
different if we use discont bace

groups
.

Exam question
!
Why do we include n-1 dummies when we have

?
n
Categories
Hint : We include n-1 dummies because including dummies will let
a
from
fall
dummy and his
the will
us into variable trap prevent us

the because
estimating coefficients including all dummies will

result in
perfect Golliverity .

This
Dummy varible trap : is when
you
include the same number
of
dummies as
categories .

Example 2

Earnings
= Po + B , Gender +
P2Pretertiary +
Potertiary
+ B42 +
p
+ ea

Reference/base : Females with no education

group
Since makes
we r e
using

Bo P
females with eduction
Earrings + +
Exp
i

no =
. e .

Males with Pretertiary a females wit no eduction :

Earnings =
Po +
B1 + Be + B + E+
P

Always Compare reac ts to

reference group

Caleb Z
Important use
of dummy variables

for for
-
Test the
Stability of Gefficients/testing Structura
Charge

Stabily of Gegents ;
Looking whter stoble time
ot Gefficients a re ove r

. We have 2 Prices in
Begora & After
both Periods
-Create
dummy
a fo r

We for Eructurch
test
change
can in
;
a

Both &
Slope Change
in
tercept
a

3
Only sope change
.

Only change
.
2
in
tercept

-
Base
Po
1

--

-Af te r

Hence for Prod 1 D = 0

,

Period 2
,
B =

Using te show test

Bo P,
Begin
+ + en - >
= r
2
.

y
= do + dim + e After
we need to find out
if
Pu = 20 and i, =
a

Lind Fadily chows

the using ze s t
;
to

Ho : Bo = - / B1 = ca

It :
Bo & puF 22
Testing Structura
Change/StaI LEy

Testing digrut Periods

-

Genficants
2
if a re ove r the

- >
Sa = Bo +, Y + u 1990 - 2000

th > 2001 2020

47 +
-

57 a
-

= 20 +
,

Introduce
during digmentiate
to btwn Period 1 & 2
-

Hand
Estime or e
equation
for bot Periods (merge
-

and introduce the

clummy

Represet (1990-2000) by
-
L (2001 -
20201
by
! Janmug
-
Introduce a
Contact
damy &
Shope damny I
for both
you'retacting

+
St Po B 47 22Diz
&
=
+ C , D + + u
,

whx = 0 & 22 = 0

both (no
change
the
periods a re
equal
st r u c t u r a l

Take I from unrestricted model

 Note

for chummins
,
the
beggcets a re alled differential Stope Cogjcets
If Produced >
hypothes
mull
reject
-

55 Prac
8 if < = and = 0 .

09 193)

Puce > & .: We

accept null

Introducing the Kategories

~

same number dummies as with an

of
intercept present in the model , there will be ex a c t linear re l a t i o n s

(ie
among
regressors
. Multicoline -

ty) .

Multicollinarity prevents from

estimating
the begficants
-

us

its is kown
dummy variable
trap
-

as the .

If want to introduce some number chimmies as categories ,

you of
drop the intercept
 Caleb Z

Model Errors
Specification
ky assumption
is that the mode is
Gorety Specifie ;
-

c Mode doesn't exclude core /relect varide

any
.

.
5 Doe not Include i r re l eva t va r i a b l e

.
2
Functional from va r i a b l e Switchy
is Chosen
of
2) No er ror
of measurent in
Legressor/Regressand
e .
Free of others in mode

f .

Eror ter m is al 10 22
,

Nonstochastic *
S
.

for Errors
-

checking specification

00mitting Gre vc r i e b e s

models here a re said to be under fitzel

d : -Lack data
maybe
to
of
-

Cardlessness

Lack
of Prope research
-

2 +
-

Po B +
Be By ddu +
Earnings
+
+ e
Age
+
= ,
p

it researcher estimate :

-
acruings
= Po + P ,
Age + BeSX10 + en Croedeation
S
W
nocht.
undersize

consequences depend on whath the a relations p Aton

the excluded variables a

Included various .

ie . Covariances stwn varises = 0

or o

If exchele variable is Grrelated to include

varid,
The estimated varibes w i ll be biased
Geggicient of
This bias does not clssepear as N
gets Urge
 be
beygiants
-

will also inconsistent

-
If exclded variables a re not Greeted with indibl

(5) would be base .

va r i e ste
, only the
intercept

The va r i a n c e e r ro r term would not

be properly
of
estimatel

This would head to estimation se

of
wrong of
the va r i a n c e s the executors and lead to
of

wrong hypothesis testing

Frecanty would also be

-

wrong
.

-
We can F test on
re st r i c t i o n to see exc h l d
use
if

Variables a re
important

ware Face = (Re -

RE)/ 2 om
ISSir-SSiar)/9
(1 -
RES/n -
1 SsRur/n- K

>
Whe H0 Be leaving Bu By
-

: &
= B 3 = 0 out

B2 + 0

Ramsey's Regression Specific Eror (RESET

When don't have iden the
tyie e r ro r
any
-

we of of
Po B, Be
earings +
Age +
Exper + e >
Original model
-

~ Procedure Crestricted mode

-

from E x i m a te Product the valus

mode
earings
-

, of
Reestimate the mode the
saunes/cubes
including
~

the estimatel value

of
Bo + B,
Age + B2 Exper + B3 earrins + B4 earrings + e

Earrings =

L
undestricte model
 T model becomes the restricted mode which the
original
-

mode
generated in
Step
2 is the Unrestricted model

Specification
~ Eor
-
-
Ho : Mode 1 is C o r re t (Pa =
+ = 0)
# i moder - is not Greet

We can use the F test on restrictions to Solve for

Tis

Lagrange Multiplier Test

-

used whe we know the variables

Earnings = Bo + B , Age + BrE +

per
+ en but we exchell Educ

O from sairtel model find the RSS (Seil)

,

② If daimtel iss the excutel

model is
Greety Specified ,
a voide

Lacub sound that be bundete

③
Regresshe e r ro r for m
of
the included a
exclude varicles

Anxilay
-

regression

If
large
*

log anxilary regressionl follows

n
is
,
n + R Chi
square
with of quitie
qual regressors
⑤ If alaleted Chi is sitide
squeld greate
than
Chi Squarel, no

reject the Ho .

(ie .

Orgul model is not

briety specific)
② Including irrelevant varies

Adding irrelevat varies is

overfitty .

Includy irrelevant verishe ahes not

my get
model

Seficients
-
a re unbissel a Consistent

-
Error vari an ce is GreetlySpecial
-

Gagehance Interven 2
hypothesis tearing world be velid

neck Sid
-

⑤ functional
Specfaction o
form

Obs

Caleb Z