Cointegration and Vector Error

Correction Model

05/11/2011 KNB/KU/Jaipuria
 Lady with her dog

05/11/2011 KNB/KU/Jaipuria
 Example2: Price of Wheat in
Futures and Spot Market

05/11/2011 KNB/KU/Jaipuria
 Cointegration: The Statistical
Concept
If two or more than two variables are
integrated of the same order d where d  0
and there exists
a stationary linear combinations of
these variables, the variables are said
to be cointegrated.

05/11/2011 KNB/KU/Jaipuria
 Example
Let x and y are I(1)
yt    xt  ut
If u(t) is stationary, x and y are cointegrated,
the regression is cointegration equation.
If u(t) is non-stationary, x and y are not
cointegrated and regression is spurious.

05/11/2011
 Error Correction
• If x and y are cointegrated at least one of
the following will produce a significant fit:

y t  1u t 1   t

xt   2 u t 1   t
• Granger Representation Theorem
05/11/2011 KNB/KU/Jaipuria
Engle-Granger Test for Cointegration in a
Bivariate I (1) Process
• Test whether both the variables x and y
follow I (1) process
• Estimate a regression equation using
variables at level, such as
yt    xt  ut

05/11/2011 KNB/KU/Jaipuria
 Engle–Granger Test……..
• Test whether u(t) is stationary. We can use
ADF test for this purpose. But the critical
values for this test will change. The modified
critical values of ADF-test in context to
cointegration are estimated by Mackinnon
(1991).
• If u(t) is stationary, the above regression
between x and y is a cointegrating equation. In
this situation β is called cointegration
parameter and (1, - β) is called cointegration
vector.
05/11/2011 KNB/KU/Jaipuria
 Engle-Granger Bivariate Error
Correction Model (ECM)

p p
yt  10  1ut 1    1i yt i   1 j xt  j   1t
i 1 j 1
p p
xt   2 0   2ut 1    2 i yt i    2 j xt  j   2 t
i 1 j 1

05/11/2011 KNB/KU/Jaipuria
 Johansen’s Cointegration Test
• Based on the evaluation of the following VECM
Δx t  Πx t 1  Γ1Δx t 1  Γ 2 Δx t  2 .........Γ 0 Dt  ε t

(Note he similarity with ADF test)

Δ xt is stationary but xt is not stationary, therefore

the equation is not balanced unless Π is of
reduced rank r<k.

05/11/2011 KNB/KU/Jaipuria