FAKULTAS EKONOMI & BISNIS

UNIVERSITAS AIRLANGGA

Cointegration and Error Correction Model (ECM)

LAB. EKONOMETRIKA II
DEPARTEMEN ILMU EKONOMI
FAKULTAS EKONOMI DAN BISNIS

1
Introduction
ECM was first used by Sargan and later popularized by Engle and Granger
corrects for disequilibrium.

ECM is used when the variables in the model are not stationary in the level I(0)
and cointegrated.

The evidence shows that regression a non stationary variable on other non
stationary variables might lead to spurious regression.

When the variables are stationary (let say) in 1st difference, I(1), these variables
probably have cointegration.

The Granger representation theorem, states that if two variables Y and X are
cointegrated, the relationship between the two can be expressed as ECM.

If the variables are cointegrated, there is long run relationship or equilibrium.

2
PEMILIHAN MODEL
Uji Stasioneritas

Stasioner pada
Stasioner pada
tingkat first
tingkat level
difference

OLS order level

Uji
Kointegrasi

Tidak
Terkointegrasi
terkointegrasi

Unrestricted Error
ECM Correction Model OLS 1st Difference
(ECM)
Steps in the ECM analysis
1. Formulating the model
2. Checking the stationarity each variable
3. Engle-Granger 2 steps Cointegration
4. Estimating ECM
5. Model diagnostic

4
 1 Formulating the model
Non stationary 𝑙𝑛𝑚𝑡 = 𝛽0 + 𝛽1 𝑙𝑛𝑦𝑡 + 𝛽2 𝑙𝑛𝑟𝑡 + 𝜀𝑡
𝜖𝑡 = 𝑙𝑛𝑚𝑡 − 𝛽0 − 𝛽1 𝑙𝑛𝑦𝑡 − 𝛽2 𝑙𝑛𝑟𝑡

∆𝑙𝑛𝑚𝑡 = 𝛽0 + 𝛽1 ∆𝑙𝑛𝑦𝑡 + 𝛽2 ∆𝑙𝑛𝑟𝑡 + 𝛽3 𝜀𝑡−1 + 𝑣𝑡

Speed of adjustment towards

the long run equilibrium
−1 < 𝛽3 < 0

2 Stationarity test
• Plotting the variables
• Testing the stationarity
5
 10.6 tsline lnm tsline lnr

3
2.8
10.4

2.6
lnm

lnr
2.4
10.2

2.2
2
10

1979q1 1981q3 1984q1 1986q3 1989q1 1979q1 1981q3 1984q1 1986q3 1989q1
t t

tsline lny
13
12.9
lny
12.8
12.7

1979q1 1981q3 1984q1 1986q3 1989q1

t
6
 Hipotesis:
H0: δ = 0, terdapat unit root/data tidak stasioner
H1: δ ≠ 0, tidak terdapat unit root/data stasioner

. dfuller lnm, trend

Dickey-Fuller test for unit root Number of obs = 39

Interpolated Dickey-Fuller
Test 1% Critical 5% Critical 10% Critical
Statistic Value Value Value

Z(t) -1.538 -4.251 -3.544 -3.206

MacKinnon approximate p-value for Z(t) = 0.8156

. dfuller lnr, drift

Dickey-Fuller test for unit root Number of obs = 39

Z(t) has t-distribution

Test 1% Critical 5% Critical 10% Critical
Statistic Value Value Value

Z(t) -1.580 -2.431 -1.687 -1.305

p-value for Z(t) = 0.0614

. dfuller lny, trend

Dickey-Fuller test for unit root Number of obs = 39

Interpolated Dickey-Fuller
Test 1% Critical 5% Critical 10% Critical
Statistic Value Value Value

Z(t) -1.103 -4.251 -3.544 -3.206

MacKinnon approximate p-value for Z(t) = 0.9286

7
 tsline dlnm, yline(0) tsline dlnr, yline(0)
.06

.4
.04

.2
.02
dlnm

dlnr

0
0

-.2
-.02

-.4
1979q1 1981q3 1984q1 1986q3 1989q1 1979q1 1981q3 1984q1 1986q3 1989q1
t t

tsline dlny, yline(0)

.03
.02
.01
dlny

0
-.01

1979q1 1981q3 1984q1 1986q3 1989q1

t 8
. dfuller dlnm

Dickey-Fuller test for unit root Number of obs = 38

Interpolated Dickey-Fuller
Test 1% Critical 5% Critical 10% Critical
Statistic Value Value Value

Z(t) -5.078 -3.662 -2.964 -2.614

MacKinnon approximate p-value for Z(t) = 0.0000

. dfuller dlnr

Dickey-Fuller test for unit root Number of obs = 38

Interpolated Dickey-Fuller
Test 1% Critical 5% Critical 10% Critical
Statistic Value Value Value

Z(t) -5.283 -3.662 -2.964 -2.614

MacKinnon approximate p-value for Z(t) = 0.0000

. dfuller dlny

Dickey-Fuller test for unit root Number of obs = 38

Interpolated Dickey-Fuller
Test 1% Critical 5% Critical 10% Critical
Statistic Value Value Value

Z(t) -3.246 -3.662 -2.964 -2.614

9
MacKinnon approximate p-value for Z(t) = 0.0175
 3 Cointegration

• Hubungan jangka panjang variabel-variabel yang tidak stasioner

dalam suatu model dapat dilihat melalui kombinasi linier.
• Jika data kedua variabel mengandung unsur unit root atau
dengan kata lain tidak stasioner, kombinasi linier kedua variabel
mungkin saja menunjukkan hal sebaliknya. Error term dalam
persamaan regresi time series merupakan kombinasi linier.
• jika variabel gangguan tidak mengandung unit root (stasioner)
pada tingkat I(0), maka kedua variabel time series di atas
terkointegrasi
• Pengujian kointegrasi dalam prosedur regresi time series
bertujuan untuk mengetahui ada atau tidaknya keseimbangan
model dalam jangka panjang.
• Uji kointegrasi hanya bisa dilakukan ketika data yang digunakan
dalam model regresi berintegrasi dalam derajat yang sama.
• Uji Kointegrasi  Augmented Engle Granger
 UJI KOINTEGRASI Engle-Granger
3
• 2-Stage Tes Kointegrasi Engle-Granger
– Langkah 1: Estimate the long run model (OLS)

𝑙𝑛𝑚𝑡 = 𝛽0 + 𝛽1 𝑙𝑛𝑦𝑡 + 𝛽2 𝑙𝑛𝑟𝑡 + 𝑢𝑡

– Langkah 2: Predict the error term and test the residual

stationarity
ut   ut 1   t Uji Unit Root ADF
– Hipotesis:
H0 :  0 If H0 is rejected ut ~ I (0)
H1 :   0
12
. reg lnm lnr lny

Source SS df MS Number of obs = 40

F( 2, 37) = 379.63
Model .927618106 2 .463809053 Prob > F = 0.0000
Residual .045203932 37 .001221728 R-squared = 0.9535
Adj R-squared = 0.9510
Total .972822038 39 .024944155 Root MSE = .03495

lnm Coef. Std. Err. t P>|t| [95% Conf. Interval]

lnr -.0666048 .0278231 -2.39 0.022 -.1229797 -.0102298

lny 1.492706 .072838 20.49 0.000 1.345122 1.640289
_cons -8.750127 .9771849 -8.95 0.000 -10.73009 -6.770163

13
 Error Correction Model (ECM)
4
• Terkointegrasinya variabel-variabel time series dalam jangka
panjang meskipun variabel-variabel tersebut tidak stasioner
dalam tingkat I(0) menunjukkan bahwa telah terjadi
disequilibrium (ketidakseimbangan) dalam jangka pendek
• diperlukan adanya penyesuaian (adjustment) untuk mengoreksi
perbedaan atau ketidakseimbangan dalam jangka pendek
tersebut  Error Correction Mechanism
ECM valid karena lagged ECT signifikan dan bernilai negatif

Interpretasi ECT  Kecepatan penyesuaian antara jangka panjang

dan jangka pendek sebesar 18 persen
15
1 2 3 4
Terima Kasih

16