Cheat Sheet for the Exam:

Econometric Methods

A-, B- and C-Assumptions

Assumption A1: (The true model) The econometric model does not lack any
relevant exogenous variables and the exogenous variables used are not
irrelevant.
Assumption A2 (Linear in parameters) The true relationship between X and
y is linear.
Assumption A3 (Constant parameters) The parameters (, 1 , 2 , . . . , K )
are constant for all T observations (xt , yt ).
Assumption B1 E(u) = o
Assumption B2 var(ut ) = 2 , f ort = 1, 2, ..., T. or V(u) = 2 IT
Assumption B3 cov(ut , us ) = 0 , for all t 6= s where t = 1, 2, ..., T und s =
1, 2, .., T. or V(u) = 2 IT
Assumption B4 The disturbance ut is normally distributed.
u N (E(u) , V(u))
Assumption C1 (Fixed exogenous variables) None of the elements of the matrix X (the exogenous variables x1t , x2t , . . . , xKt ) is a random variable,
but can be controlled as in an experiment.
Assumption C2 (Free from perfect multicollinearity) The matrix X has full
rank, i.e. all column vectors are linearly independent. So there is no
linear relationship between the columns or in other words, there is no
linear relationship between the exogenous variables where at least one
parameter k 6= 0.

Simple Regression Model

Estimation
b

b = y x
b = Sxy /Sxx

Syy
Sxx
Sxy

(yt y) (yt y)

(xt x) (xt x)

(xt x) (yt y)

Syy = Subub + Sybyb

R2 = Sybyb/Syy
E (b
) =
 
b
E
=

b2 = Subub /(T 2)
h

i
b ; b + ta/2 se(
b
b ta/2 se(
b )
b )



b ta/2 se(b
b ) ;
b + ta/2 se(b
b )

Hypothesis Testing (two sided t-test)



Pr ta/2 t ta/2
= 1a
where

t =

.
b
(b q) se(
b )

Multiple Regression Analysis

Estimation
y = X + u ,
yt = + 1 x1t + 2 x2t + ... + K xKt + ut .
1
b = (X0 X) X0 y .

b2 = Subub /(T K 1)
Variance-covariance matrix of the error

var(u1 )
cov(u1 , u2 )
cov(u2 , u1 )
var(u2 )

V(u) = E[uu0 ] =
..
..

.
.
cov(uT , u1 ) cov(uT , u2 )
R2 =

terms

..
.

cov(u1 , uT )
cov(u2 , uT )
..
.

var(uT )

= 2 IT

Sybyb
Syy Subub
=
Syy
Syy

E(b1 ) =
E(b2 ) =
E(b
)

1
2

Variance-covariance matrix of the estimators

var(b
)
cov(b
, b1 ) cov(b
, bK )

b
b
b
b)
var(1 )
cov(1 , bK )
cov(1 ,
b =
C()
.
.
..
..

..
..
.

.
b
b
b
cov(K ,
b) cov(K , 1 )
var(bK )

= 2 (X0 X)1 .

Prediction
yb0 = x00 b .
h
i
1
var(b
y0 y0 ) = 2 1 + x00 (X0 X) x0 .
Hypothesis Testing (F-Test)


Sub0ub Subub /L
F =
F(L,T K1)
Subub / (T K 1)
3

Violation of Assumptions

Assumption A1:
Omitting a relevant variable:
1
b
E(1 ) = 1 + (X01 X1 ) X01 X2 2 6= 1

e2 = E[(X2 2 )2 ] + 2
Using an irrelevant variable:
b = E()
E()

e2 = 2
Adjusted coefficient of determination R

=
=

Subub /(T K 1)
Syy /(T 1)

T 1
1 1 R2
T K 1
1

Other criteria


AIC
SC
PC


Subub
2(K + 1)
= ln
+
T
T


Subub
(K + 1) ln T
= ln
+
T
T
Subub [1 + (K + 1) /T ]
=
T K 1

Other diagnostic tools: t-test, F -test, non-nested F -test, J-test.

Assumption A2:
Nonlinear regression functions
ln yt

+ ln xt + ut

(logarithmic)

yt

+ ln xt + ut

(semi logarithmic)

ln yt

+ xt + ut

(exponential)

ln yt

+ (1/xt ) + ut

(log inverse)

yt

+ (1/xt ) + ut

(inverse)

yt

+ 1 x t +

2 x2t

+ ut

(quadratic)

Zarembas Box-Cox-Test
l
where

Subub

 

T 
Subub /
y 2 
2
ln
 (1) ,
2
Subub
= Sum of residual squares of the model with ln yt

Regression Specification Error Test (RESET)

F =

(Sub0 ub Subub ) /L

Subub / (T K 1)

F(L,T K 1)

Another diagnostic tool: R2

Assumption A3:
Diagnostic tools: F -test, t, t-test
Prognostic Chow-test:
F =

(Subub SubI ub )/TII

.
SubI ub /(TI K 1)

Assumption B2:
Goldfeld-Quandt Test
F

SuII
/(TII K 1)
bu
b
,
I
Sub ub /(TI K 1)

where SubI ub and SubIIub are the sum of residual squares of groups I and II.
White-Test
R2 T 2(v) ,
Breusch-Pagan-Test
gt =

u
b2t

bt2

BP =

Sgbgb
2

or alternatively
R2 T 2(v) ,
where =Number of slope parameters of the auxiliary model and R2 of the
auxiliary model.
5

Assumption B3:
AR(1)-Process :
= ut1 + et ,
1 < < 1

X
E(ut ) =
j E(etj ) = 0
ut

j=0

var(ut )
cov(ut , ut )

e2
2
1 2


e2

= 2 6= 0
=
1 2

Estimator for
PT
u
bt1 u
bt
.
b = Pt=2
T
2
u
b
t=2 t1
Durbin-Watson Test
PT

(b
ut u
bt1 )
PT
b2t
t=1 u
2(1 b)
t=2

d =

Durbins h-Test
s
h = b

T
1 T vd
ar(b2 )

Breusch-Godfrey-Test
BG = T R2 2(K)

Panel Estimation

Panel Estimation
Pooled Model
yi,t = + 1 x1i,t + 2 x2i,t + ... + K xKi,t + ui,t
Fixed-Effects-Model
yi,t = i + 1 x1i,t + 2 x2i,t + ... + K xKi,t + ui,t
Random-Effects-Model
yi,t = 1 x1i,t + 2 x2i,t + ... + K xKi,t + i,t
i,t = ui,t + ai