MTU,

CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

Chapt
erThr
ee
THECLASSI
CALREGRESSI
ONANALYSI
S
TheSi
[ mpl
eLi
nearRegr
essi
onModel
]

Economi
ctheor
ies ar
e mai
nly concer
ned wi
tht
he r
elat
ionshi
ps among v
ari
ous
economi
cvar
iabl
es.Theser
elat
ionshi
ps,whenphr
asedi
nmat
hemat
icalt
erms,can
pr
edi
ctt
heef
fectofonev
ari
abl
eonanot
her
. Thef
unct
ionalr
elat
ionshi
psoft
hese
v
ari
abl
esdef
inet
hedependenceofonev
ari
abl
eupont
heot
herv
ari
abl
e(s)i
nthe
speci
fi
cfor
m. Thespeci
fi
cfunct
ionalf
ormsmaybel
i
near
,quadr
ati
c,l
ogar
it
hmi
c,
exponent
ial
,hy
per
bol
i
c,oranyot
herf
orm.

I
nthi
schapt
erweshal
lconsi
derasi
mpl
eli
nearr
egr
essi
onmodel
,i.
e.ar
elat
ionshi
p
bet
weent
wov
ari
abl
esr
elat
edi
nal
i
nearf
orm.Weshal
lfi
rstdi
scusst
woi
mpor
tant
f
ormsofr
elat
ion:st
ochast
icandnon-
stochast
ic,amongwhi
chweshal
lbeusi
ngt
he
f
ormeri
neconomet
ri
canal
ysi
s.

3.
1.St
ochast
icandNon-
stochast
icRel
ati
onshi
ps
Ar
elat
ionshi
pbet
weenXandY,char
act
eri
zedasY=f
(X)i
ssai
dtobedet
ermi
nist
icor
non-
stochast
ici
fforeachv
alueoft
hei
ndependentv
ari
abl
e(X)t
her
eisoneandonl
y
onecor
respondi
ngv
alueofdependentv
ari
abl
e(Y)
.Ont
heot
herhand,ar
elat
ionshi
p
bet
weenXandYi
ssai
dtobest
ochast
ici
fforapar
ti
cul
arv
alueofXt
her
eisawhol
e
pr
obabi
l
ist
icdi
str
ibut
ionofv
aluesofY.I
nsuchacase,f
oranygi
venv
alueofX,t
he
dependentv
ari
abl
eYassumessomespeci
fi
cval
ueonl
ywi
thsomepr
obabi
l
ity
.Let
’s
i
l
lust
rat
ethedi
sti
nct
ionbet
weenst
ochast
icandnon-
stochast
icr
elat
ionshi
pswi
tht
he
hel
pofasuppl
yfunct
ion.

Assumi
ng t
hatt
he suppl
yfora cer
tai
n commodi
ty depends on i
ts pr
ice (
other
det
ermi
nant
stakent
obeconst
ant
)andt
hef
unct
ionbei
ngl
i
near
,ther
elat
ionshi
pcanbe
putas:

Theabov
erel
ati
onshi
pbet
weenPandQi
ssucht
hatf
orapar
ti
cul
arv
alueofP,
ther
eis
onl
yonecor
respondi
ngv
alueofQ.Thi
sis,t
her
efor
e,adet
ermi
nist
ic(
non-
stochast
ic)
1
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

r
elat
ionshi
p si
nce f
oreach pr
icet
her
eisal
way
sonl
yonecor
respondi
ng quant
it
y
suppl
i
ed.Thi
simpl
i
est
hatal
lthev
ari
ati
oni
nYi
sduesol
elyt
ochangesi
nX,andt
hat
t
her
ear
enoot
herf
act
orsaf
fect
ingt
hedependentv
ari
abl
e.

I
fthi
swer
etr
ueal
lthepoi
ntsofpr
ice-
quant
it
ypai
rs,i
fpl
ott
edonat
wo-
dimensi
onal
pl
ane,woul
dfal
lonast
rai
ghtl
i
ne.Howev
er,i
fwegat
herobser
vat
ionsont
hequant
it
y
act
ual
l
ysuppl
i
edi
nthemar
ketatv
ari
ouspr
icesandwepl
ott
hem onadi
agr
am wesee
t
hatt
heydonotf
all
onast
rai
ghtl
i
ne.

Theder
ivat
ionoft
heobser
vat
ionf
rom t
hel
i
nemaybeat
tr
ibut
edt
osev
eral
fact
ors.
a.Omi
ssi
onofv
ari
abl
esf
rom t
hef
unct
ion
b.Random behav
iorofhumanbei
ngs
c.I
mper
fectspeci
fi
cat
ionoft
hemat
hemat
ical
for
m oft
hemodel
d.Er
rorofaggr
egat
ion
e.Er
rorofmeasur
ement
I
nor
dert
otakei
ntoaccountt
heabov
esour
cesofer
ror
swei
ntr
oducei
neconomet
ri
c
f
unct
ionsar
andom v
ari
abl
ewhi
chi
susual
l
ydenot
edbyt
hel
ett
er‘
u’or‘’
andi
scal
l
ed
er
rort
erm orr
andom di
stur
banceorst
ochast
ict
erm oft
hef
unct
ion,
socal
l
edbecause
ui
ssupposedt
o‘di
stur
b’t
heexactl
i
nearr
elat
ionshi
pwhi
chi
sassumedt
o exi
st
bet
weenXandY. Byi
ntr
oduci
ngt
hisr
andom v
ari
abl
eint
hef
unct
iont
hemodeli
s
r
ender
edst
ochast
icoft
hef
orm:

……………………………………………………….
(2.
2)

Thusast
ochast
icmodeli
samodeli
n whi
ch t
hedependentv
ari
abl
eisnotonl
y
det
ermi
nedbyt
heexpl
anat
oryv
ari
abl
e(s)i
ncl
udedi
nthemodelbutal
sobyot
her
s
whi
char
enoti
ncl
udedi
nthemodel
.
2
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

3.
2.Si
mpl
eLi
nearRegr
essi
onmodel
.
Theabov
est
ochast
icr
elat
ionshi
p(2.
2)wi
thoneexpl
anat
oryv
ari
abl
eiscal
l
edsi
mpl
e
l
i
nearr
egr
essi
onmodel
.
Thet
ruer
elat
ionshi
pwhi
chconnect
sthev
ari
abl
esi
nvol
vedi
sspl
i
tint
otwopar
ts:
apar
trepr
esent
edbyal
i
neandapar
trepr
esent
edbyt
her
andom t
erm‘
u’
.

Thescat
terofobser
vat
ionsr
epr
esent
sthet
ruer
elat
ionshi
pbet
weenYandX. Thel
i
ne
r
epr
esent
stheexactpar
toft
her
elat
ionshi
pandt
hedev
iat
ionoft
heobser
vat
ionf
rom
t
hel
i
ner
epr
esent
sther
andom componentoft
her
elat
ionshi
p.

Wer
eitnotf
ort
heer
ror
sint
hemodel
,wewoul
dobser
veal
lthepoi
ntsont
hel
i
ne

cor
respondi
ng t
o . Howev
erbecause oft
he r
andom

di
stur
bance,weobser
ve cor
respondi
ngt
o . Thesepoi
nts

di
ver
gef
rom t
her
egr
essi
onl
i
neby .

Thef
ir
stcomponenti
nthebr
acketi
sthepar
tofYexpl
ainedbyt
hechangesi
nXandt
he
secondi
sthepar
tofYnotexpl
ainedbyX,t
hati
stosayt
hechangei
nYi
sduet
othe

r
andom i
nfl
uenceof .

3
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

3.
2.1Assumpt
ionsoft
heCl
assi
cal
LinearSt
ochast
icRegr
essi
onModel
.
Thecl
assi
calmadei
mpor
tantassumpt
ionsi
nthei
ranal
ysi
sofr
egr
essi
on.
Themost
i
mpor
tantoft
heseassumpt
ionsar
edi
scussedbel
ow.

1.Themodel
isl
i
neari
npar
amet
ers.
Thecl
assi
calassumedt
hatt
hemodelshoul
dbel
i
neari
nthepar
amet
ersr
egar
dlessof
whet
hert
heexpl
anat
oryandt
hedependentv
ari
abl
esar
eli
nearornot
.Thi
sisbecausei
f
t
hepar
amet
ersar
enon-
li
neari
tisdi
ff
icul
ttoest
imat
ethem si
ncet
hei
rval
uei
snot
knownbuty
ouar
egi
venwi
tht
hedat
aoft
hedependentandi
ndependentv
ari
abl
e.

Exampl
e1. i
sli
neari
nbot
hpar
amet
ersandt
hev
ari
abl
es,
soi
t
Sat
isf
iest
heassumpt
ion
2. i
sli
nearonl
yint
hepar
amet
ers.Si
ncet
he
t
hecl
assi
cal
swor
ryont
hepar
amet
ers,
themodel
sat
isf
ies
t
heassumpt
ion.

 Checky
our
sel
fwhet
hert
hef
oll
owi
ngmodel
ssat
isf
ytheabov
eassumpt
ionand
gi
vey
ouranswert
oyourt
utor
.

a.

b.

2. i
sar
andom r
ealv
ari
abl
e

Thi
smeanst
hatt
hev
aluewhi
chumayassumei
nanyoneper
ioddependsonchance;
it
maybeposi
ti
ve,negat
iveorzer
o. Ev
eryv
aluehasacer
tai
npr
obabi
l
ityofbei
ng
assumedbyui
nanypar
ti
cul
ari
nst
ance.

3.Themeanv
alueoft
her
andom v
ari
abl
e(U)i
nanypar
ti
cul
arper
iodi
szer
o

4
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

Thi
smeanst
hatf
oreachv
alueofx,
ther
andom v
ari
abl
e(u)mayassumev
ari
ousv
alues,
somegr
eat
ert
hanzer
oandsomesmal
l
ert
hanzer
o,buti
fweconsi
der
edal
lthe
possi
bleandnegat
ivev
aluesofu,
foranygi
venv
alueofX,
theywoul
dhav
eonav
erage
v
alueequalt
ozer
o.I
not
herwor
dst
heposi
ti
veandnegat
ivev
aluesofucanceleach
ot
her
.

Mat
hemat
ical
l
y, ……………………………….
.….(
2.3)

4.The v
ari
ance oft
he r
andom v
ari
abl
e(U)i
s const
anti
n each per
iod (
The
assumpt
ionofhomoscedast
ici
ty)
Foral
lval
uesofX,t
heu’
swi
l
lshow t
hesamedi
sper
sionar
oundt
hei
rmean.I
n
Fi
g.2.
cthi
sassumpt
ioni
sdenot
edbyt
hef
actt
hatt
hev
aluest
hatucanassumel
i
e
wi
thi
nthesamel
i
mit
s,i
rr
espect
iveoft
hev
alueofX.For ,ucanassumeany
v
aluewi
thi
nther
angeAB;f
or ,ucanassumeanyv
aluewi
thi
nther
angeCD
whi
chi
sequal
toABandsoon.
Gr
aphi
cal
l
y;

Mat
hemat
ical
l
y;

(
Since )
.Thi
sconst
antv
ari
ancei
scal
l
ed

5
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

homoscedast
ici
tyassumpt
ionandt
heconst
antv
ari
ancei
tsel
fiscal
l
edhomoscedast
ic
v
ari
ance.

4.Ther
andom v
ari
abl
e(U)hasanor
maldi
str
ibut
ion
Thi
smeanst
hev
aluesofu(
foreachx)hav
eabel
lshapedsy
mmet
ri
caldi
str
ibut
ion
aboutt
hei
rzer
omeanandconst
antv
ari
ance ,
i.
e.

 ……………………………………….
.……2.
4

5.Ther
andom t
ermsofdi
ff
erentobser
vat
ions ar
eindependent
.(The

assumpt
ionofnoaut
ocor
rel
ati
on)
Thi
smeanst
hev
aluewhi
cht
her
andom t
erm assumedi
noneper
ioddoesnotdepend
ont
hev
aluewhi
chi
tassumedi
nanyot
herper
iod.
Al
gebr
aical
l
y,

………………………….
.….(
2.5)

6.The ar
easetoff
ixedv
aluesi
nthehy
pot
het
icalpr
ocessofr
epeat
ed
sampl
i
ngwhi
chunder
li
est
hel
i
nearr
egr
essi
onmodel
.

Thi
smeanst
hat
,int
aki
ngl
argenumberofsampl
esonYandX,
the v
aluesar
e

t
hesamei
nal
lsampl
es,butt
he v
aluesdodi
ff
erf
rom sampl
etosampl
e,and

soofcour
sedot
hev
aluesof .

7.Ther
andom v
ari
abl
e(U)i
sindependentoft
heexpl
anat
oryv
ari
abl
es.
Thi
s means t
her
eis no cor
rel
ati
on bet
ween t
he r
andom v
ari
abl
e and t
he
expl
anat
oryv
ari
abl
e.I
ftwov
ari
abl
esar
eunr
elat
edt
hei
rcov
ari
ancei
szer
o.

Hence ……………………………………….
.….
(2.
6)

Pr
oof
:-

6
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

,
giv
ent
hatt
he ar
efi
xed

.
8.Theexpl
anat
oryv
ari
abl
esar
emeasur
edwi
thouter
ror
Uabsor
bst
hei
nfl
uenceofomi
tt
edv
ari
abl
esandpossi
blyer
ror
sofmeasur
ement
i
nthey
’s.i
.e.
,wewi
l
lassumet
hatt
her
epr
essor
sar
eer
rorf
ree,whi
l
eyv
alues
mayormaynoti
ncl
udeer
ror
sofmeasur
ement
.

Wecannowuset
heabov
eassumpt
ionst
oder
ivet
hef
oll
owi
ngbasi
cconcept
s.

A.Thedependentv
ari
abl
e i
snor
mal
l
ydi
str
ibut
ed.

i
.e ………………………………(
2.7)

Pr
oof
:

Mean:

Si
nce

Var
iance:

(
since )

……………………………………….
(2.
8)

Theshapeoft
hedi
str
ibut
ionof i
sdet
ermi
nedbyt
heshapeoft
hedi
str
ibut
ionof

whi
chi
snor
malbyassumpt
ion4.Si
nce ,bei
ngconst
ant
,theydon’
taf
fectt
he

di
str
ibut
ionof .Fur
ther
mor
e,t
hev
aluesoft
heexpl
anat
oryv
ari
abl
e, ,ar
easetof

f
ixedv
aluesbyassumpt
ion5andt
her
efor
edon’
taf
fectt
heshapeoft
hedi
str
ibut
ionof

7
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

B.successi
vev
aluesoft
hedependentv
ari
abl
ear
eindependent
,i.
e

Pr
oof
:

(
Since and )

= ,
Since

(
from equat
ion(
2.5)
)

Ther
efor
e, .

3.
2.2Met
hodsofest
imat
ion
Speci
fyi
ngt
hemodel
andst
ati
ngi
tsunder
lyi
ngassumpt
ionsar
ethef
ir
stst
ageofany
economet
ri
cappl
i
cat
ion.Thenextst
epi
stheest
imat
ionoft
henumer
ical
val
uesoft
he
par
amet
ersofeconomi
crel
ati
onshi
ps.Thepar
amet
ersoft
hesi
mpl
eli
nearr
egr
essi
on
model
canbeest
imat
edbyv
ari
ousmet
hods.Thr
eeoft
hemostcommonl
yused
met
hodsar
e:
1.Or
dinar
yleastsquar
emet
hod(
OLS)
2.Maxi
mum l
i
kel
i
hoodmet
hod(
MLM)
3.Met
hodofmoment
s(MM)
But
,her
ewewi
l
ldeal
wit
htheOLSandt
heMLM met
hodsofest
imat
ion.
3.
2.2.
1Theor
dinar
yleastsquar
e(OLS)met
hod

Themodel i
scal
l
edt
het
ruer
elat
ionshi
pbet
weenYandXbecauseY

andXr
epr
esentt
hei
rrespect
ivepopul
ati
onv
alue,and ar
ecal
l
edt
het
rue
par
amet
erssi
ncet
heyar
eest
imat
edf
rom t
hepopul
ati
onv
alueofYandXButi
tis
di
ff
icul
ttoobt
aint
hepopul
ati
onv
alueof YandXbecauseoft
echni
caloreconomi
c

8
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

r
easons.Sowear
efor
cedt
otaket
hesampl
eval
ueofY andX.Thepar
amet
ers
est
imat
edf
rom t
hesampl
eval
ueofYandX ar
ecal
l
edt
heest
imat
orsoft
het
rue

par
amet
ers andar
esy
mbol
i
zedas .

Themodel ,i
scal
l
edest
imat
edr
elat
ionshi
pbet
weenYandXsi
nce

ar
eest
imat
edf
rom t
hesampl
eofYandXand r
epr
esent
sthesampl
e

count
erpar
toft
hepopul
ati
onr
andom di
stur
bance .

Est
imat
ionof byl
eastsquar
emet
hod(
OLS)orcl
assi
call
eastsquar
e(CLS)

i
nvol
vesf
indi
ngv
aluesf
ort
heest
imat
es whi
chwi
l
lmi
nimi
zet
hesum of

squar
eoft
hesquar
edr
esi
dual
s( )
.

Fr
om t
heest
imat
edr
elat
ionshi
p ,
weobt
ain:

……………………………(
2.6)

……………………….(
2.7)

Tof
indt
hev
aluesof t
hatmi
nimi
zet
hissum,
wehav
etopar
ti
all
ydi
ff
erent
iat
e

wi
thr
espectt
o andsett
hepar
ti
alder
ivat
ivesequal
tozer
o.

1.

Rear
rangi
ngt
hisexpr
essi
onwewi
l
lget
: ……(
2.9)

I
fyoudi
vi
de(
2.9)by‘
n’andr
ear
range,
weget

2.

Not
e:att
hispoi
ntt
hatt
het
ermi
nthepar
ent
hesi
sinequat
ion2.
8and2.
11i
sthe

r
esi
dual
, .Hencei
tispossi
blet
orewr
it
e(2.
8)and(
2.11)as

and .I
tfol
l
owst
hat
;

9
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

and

I
fwer
ear
rangeequat
ion(
2.11)weobt
ain;

……………………………………….
(2.
13)

Equat
ion(
2.9)and(
2.13)ar
ecal
l heNor
edt malEquat
ions.Subst
it
uti
ngt
hev
aluesof
f
rom (
2.10)t
o(2.
13)
,weget
:

2
= ( )

………………….
(2.
14)

Equat
ion(
2.14)canber
ewr
it
teni
nsomewhatdi
ff
erentwayasf
oll
ows;

Subst
it
uti
ng(
2.15)and(
2.16)i
n(2.
14)
,weget

Now,
denot
ing as ,
and as weget
;

………………………………………(
2.17)

Theexpr
essi
oni
n(2.
17)t
oest
imat
ethepar
amet
ercoef
fi
cienti
ster
medi
sthef
ormul
a

10
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

ndev
i iat
ionf
orm.
3.
2.2.
2Est
imat
ionofaf
unct
ionwi
thzer
oint
ercept

Supposei
tisdesi
redt
ofi
tthel
i
ne ,
subj
ectt
other
est
ri
cti
on To

est
imat
e ,t
hepr
obl
em i
sputi
naf
orm ofr
est
ri
ctedmi
nimi
zat
ionpr
obl
em andt
hen
Lagr
angemet
hodi
sappl
i
ed.

Wemi
nimi
ze:

Subj
ectt
o:
Thecomposi
tef
unct
iont
henbecomes

wher
e i
saLagr
angemul
ti
pli
er.

Wemi
nimi
zet
hef
unct
ionwi
thr
espectt
o

Subst
it
uti
ng(
ii
i)i
n(i
i
)andr
ear
rangi
ngweobt
ain:

…………………………………….
.(
2.18)

Thi
sfor
mul
ainv
olv heact
est ualv
alues(
obser
vat
ions)oft
hev
ari
abl
esandnott
hei
r

dev
iat
ionf
orms,
asi
nthecaseofunr
est
ri
ctedv
alueof .

3.
2.2.
3.St
ati
sti
calPr
oper
ti
esofLeastSquar
eEst
imat
ors
Ther
ear
evar
iouseconomet
ri
cmet
hodswi
thwhi
chwemayobt
aint
heest
imat
esoft
he
par
amet
ersofeconomi
crel
ati
onshi
ps.Wewoul
dli
ket
oanest
imat
etobeascl
oseas

11
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

t
hev
alueoft
het
ruepopul
ati
onpar
amet
ersi
.e.t
ovar
ywi
thi
nonl
yasmal
lrangear
ound
t
het
ruepar
amet
er.Howar
ewet
ochooseamongt
hedi
ff
erenteconomet
ri
cmet
hods,
t
heonet
hatgi
ves‘
good’
est
imat
es?Weneedsomecr
it
eri
aforj
udgi
ngt
he‘
goodness’
of
anest
imat
e.

‘
Closeness’oft
heest
imat
etot
hepopul
ati
onpar
amet
eri
smeasur
edbyt
hemeanand
v
ari
anceorst
andar
ddev
iat
i hesampl
onoft i
ngdi
str
ibut
ion oft
heest
imat
esoft
he
di
ff
erenteconomet
ri
cmet
hods.Weassumet
heusualpr
ocessofr
epeat
edsampl
i
ngi
.e.
weassumet
hatwegetav
eryl
argenumberofsampl
eseachofsi
ze‘
n’
;wecomput
ethe

est
imat
es ’
sfr
om eachsampl
e,andf
oreacheconomet
ri
cmet
hodandwef
ormt
hei
r
di
str
ibut
ion.Wenextcompar
ethemean(
expect
edv
alue)andt
hev
ari
ancesoft
hese
di
str
ibut
ions and we choose among t
he al
ter
nat
ive est
imat
es t
he one whose
di
str
ibut
ioni
sconcent
rat
edascl
oseaspossi
blear
oundt
hepopul
ati
onpar
amet
er.

PROPERTI
ESOFOLSESTI
MATORS
Thei
dealoropt
imum pr
oper
ti
est
hatt
heOLSest
imat
espossessmaybesummar
ized
bywel
l
-knownt
heor heGauss-
em knownast Mar
kovTheor
em.
St
atementoft
het
heor
em:“
Giv
ent
heassumpt
ionsoft
hecl
assi
call
i
nearr
egr
essi
on
model
,theOLSest
imat
ors,i
nthecl
assofl
i
nearandunbi
asedest
imat
ors,hav
ethe
mi
nimum v
ari
ance,
i.
e.t
heOLSest
imat
orsar
eBLUE.
Accor
ding t
othe t
heor
em,undert
he basi
c assumpt
ions oft
he cl
assi
call
i
near
r
egr
essi
onmodel
,thel
eastsquar
esest
imat
orsar
eli
near
,unbi
asedandhav
emi
nimum
v
ari
ance(
i.
e.ar
ebestofal
lli
nearunbi
asedest
imat
ors)
. Somet
imest
het
heor
em
r
efer
redast
heBLUEt
heor
em i
.e.Best
,Li
near
,andUnbi
asedEst
imat
or.Anest
imat
ori
s
cal
l
edBLUEi
f:
a.Li
near
:al
i
nearf
unct
ionoft
her
andom v
ari
abl
e,suchas,t
hedependent
v
ari
abl
eY.
b.Unbi
ased:i
tsav
erageorexpect
ed v
aluei
sequalt
othet
ruepopul
ati
on
par
amet
er.
c.Mi
nimum v
ari
ance:I
thasami
nimum v
ari
ancei
nthecl
assofl
i
nearand
unbi
asedest
imat
ors.Anunbi
asedest
imat
orwi
tht
hel
eastv
ari
ancei
sknown

12
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

asanef
fi
ci
entest
imat
or.
Accor
dingt
otheGauss-
Mar
kovt
heor
em,t
heOLSest
imat
orspossessal
ltheBLUE
pr
oper
ti
es.Thedet
ail
edpr
oofoft
hesepr
oper
ti
esar
epr
esent
edbel
ow
Dearcol
l
eaguel
etspr
ooft
hesepr
oper
ti
esonebyone.

a.Li
near
it
y:(
for )

Pr
oposi
ti
on: ar
eli
neari
nY.

Pr
oof
:Fr
om (
2.17)oft
heOLSest
imat
orof i
sgi
venby
:

(
but )

; Now,
let

i
sli
neari
nY

Checky
our
sel
fquest
ion:

Showt
hat i
sli
neari
nY?Hi
nt: .Der
ivet
hisr
elat
ionshi
pbet
ween

andY.

b.Unbi
asedness:

Pr
oposi
ti
on: ar
etheunbi
asedest
imat
orsoft
het
ruepar
amet
ers
Fr
om y
ourst
ati
sti
cscour
se,y
oumayr
ecal
lthati
f i
sanest
imat
orof t
hen

andi
f i
stheunbi
asedest
imat
orof t
henbi
as=0i
.e.

I
nourcase, ar
eest
imat
orsoft
het
ruepar
amet
ers .
Toshowt
hatt
heyar
e
t
heunbi
asedest
imat
orsoft
hei
rrespect
ivepar
amet
ersmeanst
opr
ovet
hat
:

13
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

 Pr
oof(
1):Pr
ovet
hat i
sunbi
asedi
.e.

Weknowt
hat

…………………………………………………………………(
2.20)

……………………………………………(
2.21)

Si
nce ar
efi
xed

, si
nce

Ther
efor
e, i
sunbi
asedest
imat
orof .
 Pr
oof
(2)
:pr
ovet
hat i
sunbi
asedi
.e.
:
Fr
om t
hepr
oofofl
i
near
it
ypr
oper
tyunder2.
2.2.
3(a)
,weknowt
hat
:

,Si
nce

14
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

……………………(
2.23)

i
sanunbi
asedest
imat
orof .

c.Mi
nimum v
ari
anceof
Now,wehav
etoest
abl
i
sht
hatoutoft
hecl
assofl
i
nearandunbi
asedest
imat
orsof

, possesst
hesmal
l
estsampl
i
ngv
ari
ances.Fort
his,weshal
lfi
rst

obt
ainv
ari
anceof andt
henest
abl
i
sht
hateachhast
hemi
nimum v
ari
ancei
n
compar
isonoft
hev
ari
ancesofot
herl
i
nearandunbi
asedest
imat
orsobt
ainedbyany
ot
hereconomet
ri
cmet
hodst
hanOLS.

a.Var
ianceof

……………………………………(
2.25)
Subst
it
ute(
2.22)i
n(2.
25)andweget

(
Since =0)

,andt
her
efor
e,

…………………………………………….
.(2.
26)

b.Var
ianceof

15
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

Subst
it
uti
ngequat
ion(
2.23)i
n(2.
27)
,weget

,
Since

,Si
nce

Agai
n:

…………………………………………(
2.28)

Dearst
udent
! Wehav
ecomput
edt
hev
ari
ancesOLSest
imat
ors.Now,i
tist
imet
o
checkwhet
hert
hesev
ari
ancesofOLS est
imat
orsdo possessmi
nimum v
ari
ance
pr
oper
tycompar
edt
othev
ari
ancesot
herest
imat
orsoft
het
rue ,ot
hert
han

Toest
abl
i
sht
hat possessmi
nimum v
ari
ancepr
oper
ty,wecompar
ethei
r
v
ari
anceswi
tht
hatoft
hev
ari
ancesofsomeot
heral
ter
nat
ivel
i
nearandunbi
ased
est
imat
orsof ,say and .Now,wewantt
opr
ovet
hatanyot
herl
i
near
andunbi
asedest
imat
oroft
het
ruepopul
ati
onpar
amet
erobt
ainedf
rom anyot
her
economet
ri
cmet
hodhasl
argerv
ari
ancet
hatt
hatOLSest
imat
ors.

Let
sfi
rstshowmi
nimum v
ari
anceof andt
hent
hatof .

16
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

1.Mi
nimum v
ari
anceof
Suppose: anal
ter
nat
ivel
i
nearandunbi
asedest
imat
orof and;

Let ………………………………(
2.29)

wher
e, ;but
:

Si
nce

,
si
nce

Si
nce i
sassumedt
obeanunbi
asedest
imat
or,t
henf
or i
stobeanunbi
ased

est
i orof ,
mat ther
emustbet
ruet
hat and i
ntheabov
eequat
ion.

But
,

Ther
efor
e, si
nce

Agai
n

Si
nce .

Fr
om t
hesev
alueswecandr
ive

Si
nce

Thus,
from t
heabov
ecal
cul
ati
onswecansummar
izet
hef
oll
owi
ngr
esul
ts.

Topr
ovewhet
her hasmi
nimum v
ari
anceornotl
etscomput
e t
ocompar
e

wi
th .

17
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

si
nce

But
,

Si
nce

Ther
efor
e,

Gi
vent
hatcii
sanar
bit
rar
yconst
ant
, i
saposi
ti
vei
.ei
tisgr
eat
ert
hanzer
o.Thus

.Thi
spr
ovest
hat possessesmi
nimum v
ari
ancepr
oper
ty.I
nthe
si
mil
arwaywecanpr
ovet
hatt
hel
eastsquar
eest
imat
eoft
heconst
anti
nter
cept( )
possessesmi
nimum v
ari
ance.

2.Mi
nimum Var
ianceof
Wet
akeanewest
imat
or ,whi
chweassumet
obeal
i
nearandunbi
asedest
imat
or
off
unct
ionof .Thel
eastsquar
eest
imat
or i
sgi
venby
:

Byanal
ogywi
tht
hatt
hepr
oofoft
hemi
nimum v
ari
ancepr
oper
tyof ,l
et’
suset
he
wei
ght
swi=ci+kiConsequent
ly;

Si
ncewewant t
obeonunbi
asedest
imat
oroft
het
rue ,t
hati
s, ,we

subst
it
utef
or i
n andf
indt
heexpect
edv
alueof .

18
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

For t
obeanunbi
asedest
imat
oroft
het
rue ,
thef
oll
owi
ngmusthol
d.

i
.e.
,i
f Thesecondi
ti
onsi
mpl
ythat and .

Asi
nthecaseof ,
weneedt
ocomput
e )t
ocompar
ewi
thv
ar( )

,
Since

but

Thef
ir
stt
ermi
nthebr
acketi
t ,
hence

,
Since

Ther
efor
e,wehav
epr
ovedt
hatt
hel
eastsquar
eest
imat
orsofl
i
nearr
egr
essi
onmodel
ar
ebest
,li
nearandunbi
ased(
BLU)est
imat
ors.

Thev
ari
anceoft
her
andom v
ari
abl
e(Ui)
Youmayobser
vet
hatt
hev
ari
ancesoft
heOLSest
imat
esi
nvol
ve ,whi
chi
sthe
popul
ati
onv
ari
anceoft
her
andom di
stur
bancet
erm.Buti
tisdi
ff
icul
ttoobt
aint
he

19
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

popul
ati
ondat
aoft
hedi
stur
bancet
erm becauseoft
echni
calandeconomi
creasons.
Hencei
tisdi
ff
icul
ttocomput
e ;
thi
simpl
i
est
hatv
ari
ancesofOLSest
imat
esar
eal
so
di
ff
icul
ttocomput
e.Butwecancomput
ethesev
ari
ancesi
fwet
aket
heunbi
ased
est
imat
eof whi
chi
s comput
edf
rom t
hesampl
eval
ueoft
hedi
stur
bancet
erm ei
f
rom t
heexpr
essi
on:

………………………………….
.2.
30

Touse i
ntheexpr
essi
onsf
ort
hev
ari
ancesof ,wehav
etopr
ovewhet
her

i
stheunbi
asedest
imat
orof ,
i.
e.,

Topr
ovet
hiswehav
etocomput
e f
rom t
heexpr
essi
onsofY, ,

Pr
oof
:

……………………………………………………………(
2.31)

……………………………………………………………(
2.32)

Summi
ng(
2.31)wi
l
lresul
tthef
oll
owi
ngexpr
essi
on

Di
vi
dingbot
hsi
dest
heabov
eby‘
n’wi
l
lgi
veus

Put
ti
ng(
2.31)and(
2.33)t
oget
herandsubt
ract

20
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

………………………………………………(
2.34)

Fr
om (
2.34)
:

……………………………………………….
.(
2.35)

Wher
ethey
’sar
eindev
iat
ionf
orm.

Now,
wehav
etoexpr
ess i
not
herexpr
essi
onasder
ivedbel
ow.

Fr
om:

Weget
,bysubt
ract
ion

…………………………………………………….
(2.
36)

Not
ethatweassumedear
li
ert
hat, ,i
.ei
ntaki
ngav
eryl
argenumbersampl
es
weexpectUt
ohav
eameanv
alueofzer
o,buti
nanypar
ti
cul
arsi
ngl
esampl
e i
snot
necessar
il
yzer
o.
Si
mil
arl
y:Fr
om;

Weget
,bysubt
ract
ion

…………………………………………………………….
(2.
37)
Subst
it
uti
ng(
2.36)and(
2.37)i
n(2.
35)weget

Thesummat
ionov
ert
hensampl
eval
uesoft
hesquar
esoft
her
esi
dual
sov
ert
he‘
n’
sampl
esy
iel
ds:

21
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

Taki
ngex
pect
edv
alueswehav
e:

……………(
2.38)

Ther
ighthandsi
det
ermsof(
2.38)
mayber
ear
rangedasf
oll
ows

a.

si
nce

…………………………………………….
.(
2.39)

b.

Gi
ven t
hat t
he X’
s ar
e f
ixed i
n al
l sampl
es and we know t
hat

Hence

……………………………………………(
2.40)

c.-
2

=-
2

22
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

Butf
rom (
2.22), andsubst
it
utei
tint
heabov
eexpr
essi
on,
wewi
l
l

get
:

-
2

=-
2 ,
si
nce

……………………………………

……………….
(2.
41)
Consequent
ly,Equat
ion(
2.38)canbewr
it
teni
nter
msof(
2.39)
,(2.
40)and(
2.41)as

f
oll
ows: ………………………….
(2.
42)

Fr
om whi
chweget

……………………………………………….
.(
2.43)

Si
nce

Theconcl
usi
ont
hatwecandr
ivef
rom t
heabov
epr
oofi
sthatwecansubst
it
ute

f
or( ) i
nthev
ari
anceexpr
essi
onof ,si
nce .

Hencet
hef
ormul
aofv
ari
anceof becomes;

23
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

= ……………………………………(
2.44)

……………………………(
2.45)Not
e: canbe

comput
edas .

Donotwor
ryaboutt
heder
ivat
ionoft
hisexpr
essi
on!wewi
l
lper
for
mtheder
ivat
ionofi
t
i
noursubsequentsubt
opi
c.

3.
2.2.
4.St
ati
sti
calt
estofSi
gni
fi
canceoft
heOLSEst
imat
ors
(
Fir
stOr
dert
est
s)
Af
tert
heest
imat
ionoft
hepar
amet
ersandt
hedet
ermi
nat
ionoft
hel
eastsquar
e
r
egr
essi
onl
i
ne,weneedt
o know how ‘
good’i
sthef
itoft
hisl
i
net
othesampl
e
obser
vat
ion ofY and X,t
hati
sto say we need t
o measur
ethe di
sper
sion of
obser
vat
ionsar
oundt
her
egr
essi
onl
i
ne. Thi
sknowl
edgei
sessent
ialbecauset
he
cl
osert
heobser
vat
iont
othel
i
ne,t
hebet
tert
hegoodnessoff
it
,i.
e.t
hebet
teri
sthe
expl
anat
ionoft
hev
ari
ati
onsofYbyt
hechangesi
ntheexpl
anat
oryv
ari
abl
es.
Wedi
vi
det
heav
ail
abl
ecr
it
eri
aint
othr oups:t
eegr het
heor
eti
calapr
ior
icr
it
eri
a,t
he
st
ati
sti
calcr
it
eri
a,andt
heeconomet
ri
ccr
it
eri
a.Undert
hissect
ion,ourf
ocusi
son
st
ati
sti
calcr
it
eri
a(f
ir
stor
dert
est
s).Thet
womostcommonl
yusedf
ir
stor
dert
est
sin
economet
ri
canal
ysi
sar
e:
i
.Thecoef
fi
cientofdet
ermi
nat
ion(
thesquar
eoft
hecor
rel
ati
on
coef
fi
cienti
.e.R2)
.Thi
stesti
susedf
orj
udgi
ngt
heexpl
anat
ory
poweroft
hei
ndependentv
ari
abl
e(s)
.
i
i
.Thest
andar
der
rort
est
soft
heest
imat
ors.Thi
stesti
susedf
or
j
udgi
ngt
hest
ati
sti
calr
eli
abi
l
ityoft
heest
imat
esoft
her
egr
essi
on
coef
fi
cient
s.

GOODNESSOFFI
1.TESTSOFTHE‘ T’ TH R2
WI
r
2showst
heper
cent
ageoft
otalv
ari
ati
on oft
hedependentv
ari
abl
ethatcan be
expl
ainedbyt
hechangesi
ntheexpl
anat
oryv
ari
abl
e(s)i
ncl
udedi
nthemodel
.To

24
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

el
abor
atet
hisl
et’
sdr
aw ahor
izont
all
i
necor
respondi
ngt
othemeanv
alueoft
he

dependentv
ari
abl
e (
seef
igur
e‘d’bel
ow)
.Byf
it
ti
ngt
hel
i
ne wet
ryt
o

obt
aint
heexpl
anat
ionoft
hev
ari
ati
onoft
hedependentv
ari
abl
eYpr
oducedbyt
he
changesoft
heexpl
anat
oryv
ari
abl
eX.
.
Y
Y =

X
Fi
gur
e‘d’
.Act
ualandest
imat
edv
aluesoft
hedependentv
ari
abl
eY.
Ascanbeseenf
rom f
ig.
(d)abov
e, r
epr
esent
smeasur
est
hev
ari
ati
onoft
he
sampl
eobser
vat
ionv
alueoft
hedependentv
ari
abl
ear
oundt
hemean. Howev
ert
he
v
ari
ati
oni
nYt
hatcanbeat
tri
but
edt
hei
nfl
uenceofX,(
i.
e.t
her
egr
essi
onl
i
ne)i
sgi
ven
byt
hev
ert
icaldi
stance .Thepar
toft
het
otalv
ari
ati
oni
nYabout t
hatcan’
tbe
at
tr
ibut
edt
oXi
sequal
to whi
chi
sref
err
edt
oast
her
esi
dual
var
iat
ion.
I
nsummar
y:

=dev
iat
ionoft
heobser
vat
ionYif
rom t
her
egr
essi
onl
i
ne.

=dev
iat
ionofYf
rom i
tsmean.

=dev
iat
ionoft
her
egr
essed(
predi
cted)v
alue( )f
rom t
hemean.

Now,wemaywr
it
etheobser
vedYast
hesum oft
hepr
edi
ctedv
alue( )andt
he
r
esi
dual
ter
m(ei.)
.

Fr
om equat
ion(
2.34)wecanhav
etheabov
eequat
ionbuti
ndev
iat
ionf
orm
.Bysquar
ingandsummi
ngbot
hsi
des,
weobt
aint
hef
oll
owi
ngexpr
essi
on:

25
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

But =

(
but )

………………………………………………(
2.46)

Ther
efor
e;
……………………………….
..
(2.
47)

OR,

i
.e
……………………………………….
(2.
48)
Mat
hemat
ical
l
y;t
he expl
ained v
ari
ati
on as a per
cent
age oft
he t
otalv
ari
ati
on i
s
expl
ainedas:

……………………………………….
(2.
49)

Fr
om equat
ion(
2.37)wehav
e .Squar
ingandsummi
ngbot
hsi
desgi
veus

Wecansubst
it
ute(
2.50)i
n(2.
49)andobt
ain:

…………………………………(
2.51)

26
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

,Si
nce

………………………………………(
2.52)

Compar
ing(
2.52)wi
tht
hef
ormul
aoft
hecor
rel
ati
oncoef
fi
cient
:

Y)/x2x2 = /nx2x2 = 1/
2
r=Cov(
X, /( ) ………(
2.53)
2 2
Squar
ing(
2.53)wi
l
lresul
tin: r =( )/( )
.………….
(2.
54)
Compar
ing(
2.52)and(
2.54)
,weseeexact
lyt
heexpr
essi
ons.Ther
efor
e:

2
ESS/
TSS =r

Fr
om (
2.48) ESS.HenceR2becomes;
,RSS=TSS-

………………………….
…………(
2.55)

Fr
om equat
ion(
2.55)wecandr
ive;

Thel
i
mitofR2:Thev
alueofR2f
all
sbet
weenzer
oandone.i
.e. .

I
nter
pret
atonofR2
i
Suppose ,t
hismeanst
hatt
her
egr
essi
onl
i
negi
vesagoodf
itt
otheobser
ved
dat
asi
ncet
hisl
i
neexpl
ains90%oft
het
otalv
ari
ati
onoft
heYv
aluear
oundt
hei
rmean.
Ther
emai
ning10%oft
het
otalv
ari
ati
oni
nYi
sunaccount
edf
orbyt
her
egr
essi
onl
i
ne

andi
sat
tri
but
edt
othef
act
orsi
ncl
udedi
nthedi
stur
bancev
ari
abl
e

Checky
our
sel
fquest
ion:
a.Showt
hat .
b.Showt
hatt
hesquar
eoft
hecoef
fi
cientofcor
rel
ati
oni
sequal
toESS/
TSS.

Exer
cise:

Suppose i
sthecor
rel
ati
oncoef
fi
cientbet
weenYandXandi
sgi
veby
:

27
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

Andl
et t
hesquar
eoft
hecor
rel
ati
oncoef
fi
cientbet
weenYand ,andi
sgi
ven

by
:

Showt
hat
: i
) i
i
)

2.TESTI
NGTHESI
GNI
FICANCEOFOLSPARAMETERS
Tot
estt
hesi
gni
fi
canceoft
heOLSpar
amet
erest
imat
orsweneedt
hef
oll
owi
ng:
 Var
ianceoft
hepar
amet
erest
imat
ors
 Unbi
asedest
imat
orof
 Theassumpt
ionofnor
mal
i
tyoft
hedi
str
ibut
ionofer
rort
erm.

Wehav
eal
readyder
ivedt
hat
:

Fort
hepur
poseofest
imat
ionoft
hepar
amet
erst
heassumpt
ionofnor
mal
i
tyi
snot
used,butweuset
hisassumpt
iont
otestt
hesi
gni
fi
canceoft
hepar
amet
erest
imat
ors;
becauset
het
est
ing met
hodsorpr
ocedur
esar
ebased on t
heassumpt
ion oft
he
nor
mal
i
tyassumpt
ionoft
hedi
stur
bancet
erm.Hencebef
orewedi
scussont
hev
ari
ous
t
est
ingmet
hodsi
tisi
mpor
tantt
oseewhet
hert
hepar
amet
ersar
enor
mal
l
ydi
str
ibut
ed
ornot
.

Wehav
eal
readyassumedt
hatt
heer
rort
ermi
snor
mal
l
ydi
str
ibut
edwi
thmeanzer
o

28
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

andv
ari
ance ,i
.e. .Si
mil
arl
y,weal
sopr
ovedt
hat .

Now,
wewantt
oshowt
hef
oll
owi
ng:

1.

2.

Toshowwhet
her ar
enor
mal
l
ydi
str
ibut
edornot
,weneedt
omakeuseofone
pr
oper
tyofnor
maldi
str
ibut
ion.“
..
..
..
..anyl
i
nearf
unct
ionofanor
mal
l
ydi
str
ibut
ed
v
ari
abl
eisi
tsel
fnor
mal
l
ydi
str
ibut
ed.
”

Si
nce ar
eli
neari
nY,
itf
oll
owst
hat

TheOLSest
imat
es ar
eobt
ainedf
rom asampl
eofobser
vat
ionsonYandX.
Si
ncesampl
i
nger
ror
sar
einev
itabl
einal
lest
imat
es,i
tisnecessar
ytoappl
ytestof
si
gni
fi
cancei
nor
dert
omeasur
ethesi
zeoft
heer
roranddet
ermi
net
hedegr
eeof
conf
idencei
nor
dert
omeasur
ethev
ali
dit
yoft
heseest
imat
es.Thi
scanbedoneby
usi
ngv
ari
oust
est
s.Themostcommononesar
e:
i
) St
andar
der
rort
esti
i
)St
udent
’st
-t
est i
i
i)Conf
idencei
nter
val

Al
loft
heset
est
ingpr
ocedur
esr
eachont
hesameconcl
usi
on.Letusnow seet
hese
t
est
ingmet
hodsonebyone.
i
) St
andar
der
rort
est

Thi
stesthel
psusdeci
dewhet
hert
heest
imat
es ar
esi
gni
fi
cant
lydi
ff
erent
f
rom z
ero,i
.e.whet
hert
hesampl
efr
om whi
cht
heyhav
ebeenest
imat
edmi
ghthav
e
comef
rom apopul
ati
onwhoset
ruepar
amet
ersar
ezer
o. .
For
mal
l
ywet
estt
henul
lhy
pot
hesi
s

29
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

agai
nstt
heal
ter
nat
ivehy
pot
hesi
s

Thest
andar
der
rort
estmaybeout
li
nedasf
oll
ows.
Fi
rst
:Comput
est
andar
der
roroft
hepar
amet
ers.

Second:compar
ethest
andar
der
ror
swi
tht
henumer
ical
val
uesof .
Deci
sionr
ule:

 I
f ,
acceptt
henul
lhy
pot
hesi
sandr
ejectt
heal
ter
nat
ivehy
pot
hesi
s.

Weconcl
udet
hat i
sst
ati
sti
cal
l
yinsi
gni
fi
cant
.

 I
f ,
rej
ectt
henul
lhy
pot
hesi
sandacceptt
heal
ter
nat
ivehy
pot
hesi
s.

Weconcl
udet
hat i
sst
ati
sti
cal
l
ysi
gni
fi
cant
.

Theaccept
anceorr
eject
ionoft
henul
lhy
pot
hesi
shasdef
ini
teeconomi
cmeani
ng.
Namel
y,t
heaccept
anceoft
henul
lhy
pot
hesi
s (
thesl
opepar
amet
eri
szer
o)
i
mpl
i
est
hatt
heexpl
anat
oryv
ari
abl
etowhi
cht
hisest
imat
erel
atesdoesnoti
nfact
i
nfl
uencet
hedependentv
ari
abl
eYandshoul
dnotbei
ncl
udedi
nthef
unct
ion,
sincet
he
conduct
edt
estpr
ovi
dedev
idencet
hatchangesi
nXl
eav
eYunaf
fect
ed.I
not
herwor
ds
accept
ance ofH0 i
mpl
i
es t
hatt
he r
elat
ion shi
p bet
ween Y and X i
sinf
act
,
i.
e.t
her
eisnor
elat
ionshi
pbet
weenXandY.
Numer
icalexampl
e: Supposet
hatf
rom asampl
eofsi
zen=30,weest
imat
ethe
f
oll
owi
ngsuppl
yfunct
ion.

Testt
hesi
gni
fi
canceoft
hesl
opepar
amet
erat5% l
evelofsi
gni
fi
canceusi
ngt
he
st
andar
der
rort
est
.

30
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

Thi
simpl
i
est
hat .Thei
mpl
i
cat
ioni
s i
sst
ati
sti
cal
l
ysi
gni
fi
cantat5%

l
evel
ofsi
gni
fi
cance.
Not
e:Thest
andar
der
rort
esti
sanappr
oxi
mat
edt
est(
whi
chi
sappr
oxi
mat
edf
rom t
hez
-
testandt
-t
est
)andi
mpl
i
esat
wot
ailt
estconduct
edat5%l
evelofsi
gni
fi
cance.
i
i
) St
udent
’st
-t
est
Li
ket
hest
andar
der
rort
est
,thi
stesti
sal
soi
mpor
tantt
otestt
hesi
gni
fi
canceoft
he
par
amet
ers.Fr
om y
ourst
ati
sti
cs,anyv
ari
abl
eXcanbet
ransf
ormedi
ntotusi
ngt
he
gener
alf
ormul
a:

,wi
thn-
1degr
eeoff
reedom.

Wher
e v
alueoft
hepopul
ati
onmean

sampl
eest
imat
eoft
hepopul
ati
onst
andar
ddev
iat
ion

Sampl
esi
ze
Wecander
ivet
het
-val
ueoft
heOLSest
imat
es

wi
thn-
kdegr
eeoff
reedom.

Wher
e:
SE=i
sst
andar
der
ror
k=numberofpar
amet
ersi
nthemodel
.

Si
ncewehav
etwopar
amet
ersi
nsi
mpl
eli
nearr
egr
essi
onwi
thi
nter
ceptdi
ff
erentf
rom
zer
o,ourdegr
eeoff
reedom i
sn-
2.Li
ket
hest
andar
der
rort
estwef
ormal
l
ytestt
he

hy
pot
hesi
s: agai
nstt
heal
ter
nat
ive f
ort
hesl
opepar
amet
er;

31
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

and agai
nstt
heal
ter
nat
ive f
ort
hei
nter
cept
.

Tounder
taket
heabov
etestwef
oll
owt
hef
oll
owi
ngst
eps.
St
ep1:Comput
et*
,whi
chi
scal
l
edt
hecomput
edv
alueoft
,byt
aki
ngt
hev
alueof i
n
t
henul
lhy
pot
hesi
s.I
nourcase ,
thent
*becomes:

St
ep2:Choosel
evelofsi
gni
fi
cance.Lev
elofsi
gni
fi
cancei
sthepr
obabi
l
ityofmaki
ng
‘
wrong’
deci
sion,
i.
e.t
hepr
obabi
l
ityofr
eject
ingt
hehy
pot
hesi
swheni
tisact
ual
l
ytr
ueor
t
hepr
obabi
l
ityofcommi
tt
ingat
ypeIer
ror
.Iti
scust
omar
yineconomet
ri
cresear
cht
o
chooset
he5%ort
he1%l
evelofsi
gni
fi
cance.Thi
smeanst
hati
nmaki
ngourdeci
sion
weal
l
ow (
tol
erat
e)f
ivet
imesoutofahundr
edt
obe‘
wrong’i
.e.r
ejectt
hehy
pot
hesi
s
wheni
tisact
ual
l
ytr
ue.
St
ep3:Checkwhet
hert
her
eisonet
ailt
estort
wot
ailt
est
.Ift
hei
nequal
i
tysi
gni
nthe
al
ter
nat
ivehy
pot
hesi
sis ,
theni
timpl
i
esat
wot
ailt
estanddi
vi
det
hechosenl
evelof
si
gni
fi
cancebyt
wo;deci
det
hecr
it
icalr
ejoi
norcr
it
icalv
alueoftcal
l
edt
c.Buti
fthe
i
nequal
i
tysi
gni
sei
ther>or<t
heni
tindi
cat
esonet
ail
testandt
her
eisnoneedt
odi
vi
de
t
hechosenl
evel
ofsi
gni
fi
cancebyt
wot
oobt
aint
hecr
it
ical
val
ueoft
ofr
om t
het
-t
abl
e.

Exampl
e:

I
fwehav
e

agai
nst
:

Thent
hisi
sat
wot
ailt
est
.Ift
hel
evelofsi
gni
fi
cancei
s5%,di
vi
dei
tbyt
wot
oobt
ain
cr
it
ical
val
ueoftf
rom t
het
-t
abl
e.

St
ep4:Obt
aincr
it
icalv
alueoft
,cal
l
edt
cat andn-
2degr
eeoff
reedom f
ort
wot
ail
t
est
.
St
ep5:Compar
et*(
thecomput
edv
alueoft
)andt
c(cr
it
ical
val
ueoft
)

 I
ft*
>tc,
rej
ectH0andacceptH1.Theconcl
usi
oni
s i
sst
ati
sti
cal
l
ysi
gni
fi
cant
.

32
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

 I
ft*
<tc ,acceptH0 and r
ejectH1. The concl
usi
on i
s i
s st
ati
sti
cal
l
y
i
nsi
gni
fi
cant
.

Numer
ical
Exampl
e:
Supposet
hatf
rom asampl
esi
zen=20weest
imat
ethef
oll
owi
ngconsumpt
ionf
unct
ion:

Thev
aluesi
nthebr
acket
sar
est
andar
der
ror
s.Wewantt
otestt
henul
lhy
pot
hesi
s:

agai
nstt
heal
ter
nat
ive usi
ngt
het
-t
estat5%l
evel
ofsi
gni
fi
cance.

a.t
het
-val
uef
ort
het
estst
ati
sti
cis:

b.Si
ncet
heal
ter
nat
ivehy
pot
hesi
s(H1)i
sst
atedbyi
nequal
i
tysi
gn( ),
i
tisat
wo
t
ailt
est
,hencewedi
vi
de t
oobt
aint
hecr
it
icalv
alueof‘
t’at
=0.
025and18degr
eeoff
reedom (
df)i
.e.(
n-2=20-
2).Fr
om t
he
t
-t
abl
e‘t
c’at0.
025l
evel
ofsi
gni
fi
canceand18dfi
s2.
10.

c.Si
ncet
*=3.
3andt
c=2.
1,t
*>t
c.I
timpl
i
est
hat i
sst
ati
sti
cal
l
ysi
gni
fi
cant
.

i
i
i) Conf
idencei
nter
val

Rej
ect
ionoft
henul
lhy
pot
hesi
sdoesn’
tmeant
hatourest
imat
e i
sthecor
rect
est
imat
eoft
het
ruepopul
ati
onpar
amet
er .I
tsi
mpl
ymeanst
hatourest
imat
e
comesf
rom asampl
edr
awnf
rom apopul
ati
onwhosepar
amet
er i
sdi
ff
erentf
rom
zer
o.

I
nor
dert
odef
inehow cl
oset
heest
imat
etot
het
ruepar
amet
er,wemustconst
ruct
conf
idencei
nter
valf
ort
het
ruepar
amet
er,i
not
herwor
dswemustest
abl
i
shl
i
mit
ing
v
aluesar
oundt
heest
imat
ewi
thi
nwhi
cht
het
ruepar
amet
eri
sexpect
edt
oli
ewi
thi
na
cer
tai
n“degr
eeofconf
idence”
.Int
hisr
espectwesayt
hatwi
thagi
venpr
obabi
l
ityt
he
popul
ati
onpar
amet
erwi
l
lbewi
thi
nthedef
inedconf
idencei
nter
val
(conf
idencel
i
mit
s).

33
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

We choose a pr
obabi
l
ityi
n adv
ance and r
efert
oitas conf
idence l
evel(
int
erv
al
coef
fi
cient
).I
tiscust
omar
il
yineconomet
ri
cst
ochooset
he95%conf
idencel
evel
.Thi
s
meanst
hati
nrepeat
edsampl
i
ngt
heconf
idencel
i
mit
s,comput
edf
rom t
hesampl
e,
woul
dincl
udet
het
ruepopul
ati
onpar
amet
eri
n95%oft
hecases.I
ntheot
her5%oft
he
casest
hepopul
ati
onpar
amet
erwi
l
lfal
lout
sidet
heconf
idencei
nter
val
.
I
nat
wo-
tai
ltestatl
evelofsi
gni
fi
cance,t
hepr
obabi
l
ityofobt
aini
ngt
hespeci
fi
ct-
v
alueei
ther–t
cort
cis atn-
2degr
eeoff
reedom.Thepr
obabi
l
ityofobt
aini
ngany

v
alueoftwhi
chi
sequal
to atn-
2degr
eeoff
reedom i
s .

i
.e. …………………………………………(
2.57)

but …………………………………………………….
(2.
58)

Subst
it
ute(
2.58)i
n(2.
57)weobt
aint
hef
oll
owi
ngexpr
essi
on.

……………………………………….
.(
2.59)

Thel
i
mitwi
thi
nwhi
cht
het
rue l
i
esat degr
eeofconf
idencei
s:

;
wher
e i
sthecr
it
icalv
alueoftat conf
idencei
nter
valand

n-
2degr
eeoff
reedom.
Thet
estpr
ocedur
eisout
li
nedasf
oll
ows.

Deci
sionr
ule: I
fthehy
pot
hesi
zedv
alueof i
nthenul
lhy
pot
hesi
siswi
thi
nthe

34
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

conf
idencei
nter
val
,acceptH0 andr
ejectH1.Thei
mpl
i
cat
ioni
sthat i
sst
ati
sti
cal
l
y
i
nsi
gni
fi
cant
;whi
l
eift
hehy
pot
hesi
zedv
alueof i
nthenul
lhy
pot
hesi
sisout
sidet
he

l
i
mit
,rej
ectH0andacceptH1.Thi
sindi
cat
es i
sst
ati
sti
cal
l
ysi
gni
fi
cant
.
Numer
ical
Exampl
e:
Suppose we hav
e est
imat
ed t
he f
oll
owi
ng r
egr
essi
on l
i
ne f
rom a sampl
e of20
obser
vat
ions.

Thev
aluesi
nthebr
acketar
est
andar
der
ror
s.
a.Const
ruct95%conf
idencei
nter
val
fort
hesl
opeofpar
amet
er
b.Testt
he si
gni
fi
cance oft
he sl
ope par
amet
erusi
ng const
ruct
ed conf
idence
i
nter
val
.
Sol
uti
on:
a.Thel
i
mitwi
thi
nwhi
cht
het
rue l
i
esat95%conf
idencei
nter
val
is:

at0.
025l
evel
ofsi
gni
fi
canceand18degr
eeoff
reedom i
s2.
10.

Theconf
idencei
nter
val
is:
(
1.09,
4.67)
b.Thev
alueof i
nthenul
lhy
pot
hesi
siszer
owhi
chi
mpl
i
esi
tisoutsi
det
he
conf
idencei
nter
val
.Hence i
sst
ati
sti
cal
l
ysi
gni
fi
cant
.

2.
2.3Repor
ti
ngt
heResul
tsofRegr
essi
onAnal
ysi
s
Ther
esul
tsoft
her
egr
essi
onanal
ysi
sder
ivedar
erepor
tedi
nconv
ent
ionalf
ormat
s.I
tis
notsuf
fi
cientmer
elyt
orepor
ttheest
imat
esof ’
s.I
npr
act
icewer
epor
tregr
essi
on
coef
fi
cient
stoget
herwi
tht
hei
rst
andar
der
ror
sandt
hev
alueofR2.I
thasbecome
35
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

cust
omar
yto pr
esentt
he est
imat
ed equat
ions wi
th st
andar
d er
ror
s pl
aced i
n
par
ent
hesi
s bel
ow t
he est
imat
ed par
amet
er v
alues.Somet
imes,t
he est
imat
ed
coef
fi
cient
s,t
hecor
respondi
ngst
andar
der
ror
s,t
hep-
val
ues,
andsomeot
heri
ndi
cat
ors
ar
epr
esent
edi
ntabul
arf
orm.
Theser
esul
tsar
esuppl edbyR2on(t
ement other
ightsi
deoft
her
egr
essi
onequat
ion)
.

Exampl
e: , R2 =0.
93. Thenumber
sint
hepar
ent
hesi
s

bel
ow t
hepar
amet
erest
imat
esar
ethest
andar
der
ror
s.Someeconomet
ri
ciansr
epor
t
t
het
-val
uesoft
heest
imat
edcoef
fi
cient
sinpl
aceoft
hest
andar
der
ror
s.
Rev
iewQuest
ions

Rev
iewQuest
ions
1.Economet
ri
csdeal
swi
tht
hemeasur
ementofeconomi
crel
ati
onshi
pswhi
char
e
st
ochast
icorr
andom.Thesi
mpl
estf
orm ofeconomi
crel
ati
onshi
psbet
weent
wo
v
ari
abl
esXandYcanber
epr
esent
edby
:

; wher
e ar
eregr
essi
onpar
amet
ers

and t
hest
ochast
icdi
stur
bancet
erm

Whatar
ether
easonsf
ort
hei
nser
ti
onofU-
ter
mint
hemodel
?
2.Thef
oll
owi
ngdat
aref
erst
othedemandf
ormoney(
M)andt
her
ateofi
nter
est(
R)
i
nforei
ghtdi
ff
erenteconomi
cs:
M(
Inbi
l
li
ons) 56 50 46 30 20 35 37 61
R% 6.
3 4.
6 5.
1 7.
3 8.
9 5.
3 6.
7 3.
5

a.Assumi
ng a r
elat
ionshi
p ,obt
aint
he OLS est
imat
ors of

b.Cal
cul
atet
hecoef
fi
cientofdet
ermi
nat
ionf
ort
hedat
aandi
nter
preti
tsv
alue
c.I
fina9th economyt
her
ateofi
nter
esti
sR=8.
1,pr
edi
ctt
hedemandf
or
money
(M)i
nthi
seconomy
.
3.Thef
oll
owi
ngdat
aref
erst
othepr
iceofagood‘
P’andt
hequant
it
yoft
hegood
suppl
i
ed,
‘S’
.
P 2 7 5 1 4 8 2 8

36
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

S 15 41 32 9 28 43 17 40
a.Est
imat
ethel
i
nearr
egr
essi
onl
i
ne

b.Est
imat
ethest
andar
der
ror
sof
c.Testt
hehy
pot
hesi
sthatpr
icei
nfl
uencessuppl
y
d.Obt
aina95%conf
idencei
nter
val
for
4.Thef
oll
owi
ngr
esul
tshav
ebeenobt
ainedf
rom asi
mpl
eof11obser
vat
ionsont
he
v
aluesofsal
es(
Y)ofaf
ir
m andt
hecor
respondi
ngpr
ices(
X).

i
) Est
imat
ether
egr
essi
onl
i
neofsal
eonpr
iceandi
nter
prett
her
esul
ts
i
i
) Whati
sthepar
toft
hev
ari
ati
oni
nsal
eswhi
chi
snotexpl
ainedbyt
he
r
egr
essi
onl
i
ne?
i
i
i) Est
imat
ethepr
iceel
ast
ici
tyofsal
es.
5.Thef
oll
owi
ngt
abl
eincl
udest
heGNP(
X)andt
hedemandf
orf
ood(
Y)f
ora
count
ryov
ert
eny
ear
sper
iod.
Year 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989
Y 6 7 8 10 8 9 10 9 11 10
X 50 52 55 59 57 58 62 65 68 70
a.Est
imat
ethef
oodf
unct
ion
b.Comput
ethe coef
fi
cientofdet
ermi
nat
ion and f
ind t
he expl
ained and
unexpl
ainedv
ari
ati
oni
nthef
oodexpendi
tur
e.
c.Comput
ethest
andar
der
roroft
her
egr
essi
oncoef
fi
cient
sandconductt
estof
si
gni
fi
canceatt
he5%l
evel
ofsi
gni
fi
cance.
6.A sampl
e of 20 obser
vat
ions cor
respondi
ng t
o t
he r
egr
essi
on model

gav
ethef
oll
owi
ngdat
a.

37
 MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

a.Est
imat
e
b.Cal
cul
atet
hev
ari
anceofourest
imat
es
c.Est
imat
ethecondi
ti
onalmeanofYcor
respondi
ngt
oav
alueofXf
ixedat
X=10.
7.Supposet
hatar
esear
cherest
imat
esaconsumpt
ionsf
unct
ionandobt
ainst
he
f
oll
owi
ngr
esul
ts:

wher
eC=Consumpt
ion, Yd=di
sposabl
eincome,andnumber
sint
hepar
ent
hesi
sar
e
t
he‘
t-
rat
ios’
a.Testt
hesi
gni
fi
cantofYdst
ati
sti
cal
l
yusi
ngt
-r
ati
os
b.Det
ermi
ne t
he est
imat
ed st
andar
d dev
iat
ions of t
he par
amet
er
est
imat
es
8.St
ateandpr
oveGuass-
Mar
kovt
heor
em
9.Gi
vent
hemodel
:
10.

wi
thusualOLSassumpt
ions.Der
ivet
heexpr
essi
onf
ort
he

er
rorv
ari
ance.

38
MTU,
CANR,
Depar
tmentofAgr
icul
tur
alEconomi
cs

39