Econometrics (EM2008) Specification Error in The The K-Variable Model
Econometrics (EM2008) Specification Error in The The K-Variable Model
Lecture 3
Specification error in the the k-variable model
Irene Mammi
irene.mammi@unive.it
1 / 35
outline
I References:
I Johnston, J. and J. DiNardo (1997), Econometrics Methods, 4th
Edition, McGraw-Hill, New York, Chapter 4.
2 / 35
specification error
3 / 35
possible problems with u
1. ui are white noise but not normally distributed: this does not destroy
the BLUE property of OLS, but inference procedures are only
asymptotically valid
2. E(uu 0 ) = diag[σ12 · · · σn2 ]: the assumption of homoskedasticity is
violated The variance-covariance matrix for u is diagonal
3. E(ut ut −s ) 6= 0,(s 6= 0): the errors are pairwise correlated
4 / 35
possible problems with X
4. the X matrix has less than full column rank: this precludes estimation
of a unique b vector
5. nonzero correlations between the regressors and the errors
6. nonstationary variables
5 / 35
possible problems with β
6 / 35
tests of parameter constancy
7 / 35
tests of parameter constancy (cont.)
b 1 = (X 10 X 1 )−1 X 10 y 1
ŷ 2 = X 2 b 1
8 / 35
tests of parameter constancy (cont.)
d = y 2 − ŷ 2 = y 2 − X 2 b 1
d = y 2 − X 2 b 1 = u 2 − X 2 (b 1 − β )
var(d ) = E(d d 0 )
= σ2 I n2 + X 2 · var(b 1 ) · X 20
= σ2 [I n2 + X 2 (X 10 X 1 )−1 X 20 ]
9 / 35
tests of parameter constancy (cont.)
y 1 = X 1 β + u1
y 2 = X 2 α + u 2 = X 2 β + X 2 (α − β) + u 2 = X 2 β + γ + u 2
where γ = X 2 (α − β). If γ = 0, then α = β, and the coefficient
vector is constant over the estimation and forecast periods
(e 0∗ e ∗ − e 10 e 1 )/n2
F = ∼ F (n2 , n1 − k )
e 10 e 1 /(n1 − k )
11 / 35
tests of parameter constancy (cont.)
15 / 35
tests of parameter constancy (cont.)
the Ramsey RESET test
H0 : u ∼ N (0, σ2 I )
H1 : u ∼ N (µ, σ2 I ) µ 6= 0
y = Xβ + Zα + u
16 / 35
tests of parameter constancy (cont.)
Z = ŷ 2 ŷ 3 ŷ 4
0
where ŷ = Xb and ŷ 2 = ŷ 21 ŷ 22 · · · ŷ 2n etc. The first power,
17 / 35
tests of structural change
H0 : β 1 = β 2
18 / 35
tests of structural change (cont.)
19 / 35
tests of structural change (cont.)
I denoting RSS from the restricted model as e 0∗ e ∗ , the test of the null
is given by
(e 0 e ∗ − e 0 e )/k
F = ∗0 ∼ F (k, n − 2k )
e e/(n − 2k )
20 / 35
tests of structural change (cont.)
21 / 35
tests of structural change (cont.)
tests of slope coefficients
X 1∗ X 2∗
X1 = i1 X2 = i2
β1∗ = β2∗
22 / 35
tests of structural change (cont.)
I the unrestricted model is
α1
X 1∗
y1 i 0 0 α2
= 1
∗ ∗ + u
y2 0 i2 0 X2 β1
β2∗
I the restricted model is
α
X 1∗ 1
y1 i 0
= 1 α2 + u
y2 0 i2 X 2∗
β∗
I the test can be based on the RSS from these two regressions
I an alternative formulation of the unrestricted model is
α1
0 X 1∗ 0
y1
= 1
i α2 −∗ α1 + u
∗ ∗
y2 i2 i2 X2 X2 β1
β2 − β1∗
∗
X 1∗
y1 i α
= 1 +u
y2 i2 X 2∗ β∗
24 / 35
tests of structural change (cont.)
25 / 35
tests of structural change (cont.)
summary
I there is a hierarchy of the three models, namely,
X 1∗
y1 i α
I: = 1 +u Common parameters
y2 i2 X 2∗ β∗
α
X 1∗ 1
y1 i 0 Differential intercepts
II : = 1 α2 + u
y2 0 i2 X 2∗ common slope vectors
β∗
α1
X 1∗ 0
y1 i 0 α2 + u Differential intercepts
= 1
III :
y2 0 i2 0 X ∗ β∗
2 1 differential slope vectors
β2∗
26 / 35
tests of structural change (cont.)
I the test statistics for various hypotheses are as follows
RSS1 − RSS2
F = ∼ F (1, n − k − 1)
RSS2 /(n − k − 1)
27 / 35
dummy variables
examples
28 / 35
dummy variables (cont.)
29 / 35
dummy variables (cont.)
30 / 35
dummy variables (cont.)
seasonal dummies
for i = 1, . . . , 4
I for the four quarters of each year these dummies are
Q1 Q2 Q3 Q4
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
31 / 35
dummy variables (cont.)
I the relationship may then be written as
Yt = α1 Q1t + · · · + α4 Q4t + x t0 β + ut
I an alternative specification is
γ2 = α2 − α1 γ3 = α3 − α1 γ4 = α4 − α1
H0 : α1 = α2 = α3 = α4
H0 : γ2 = γ3 = γ4 = 0
32 / 35
dummy variables (cont.)
qualitative dummies
S1 = 1 if male
= 0 otherwise
and
S2 = 1 if female
= 0 otherwise
33 / 35
dummy variables (cont.)
I typical entries for the S and R dummies would be
S1 S2 R1 R2 R3
0 1 1 0 0
1 0 0 1 0
Y = α1 E1 + α2 E2 + α3 E3 + α4 E4 + u
E(Y |Ei ) = α1 i = 1, . . . , 4
Y = α1 + γ2 E2 + γ3 E3 + γ4 E4 + u
34 / 35
dummy variables (cont.)
I the γs measure the marginal increment in expected income for a
diploma over the no-diploma level
I to measure the stepwise marginal increments, let the same dummies
to have value of one if a person has the relevant diploma, irrespective
of whether she has one or more diplomas, and E1 to be the pre-high
school diploma dummy
I the equation to be fitted is
Y = α1 + δ2 E2 + δ3 E3 + δ4 E4 + u
E (Y |pre-HS diploma) = α1
E (Y |HS diploma) = α1 + δ2
E (Y |bachelor diploma) = α1 + δ2 + δ3
E (Y |graduate diploma) = α1 + δ2 + δ3 + δ4
I the δs provide estimates of the marginal increment from one level to
the next higher level
35 / 35