Journal of Physics: Conference Series

PAPER • OPEN ACCESS

Modeling Multivariate Time Series by Vector Error Correction Models

(VECM) (Study: PT Kalbe Farma Tbk. and PT Kimia Farma (Persero)
Tbk)
To cite this article: L Loves et al 2021 J. Phys.: Conf. Ser. 1751 012013

View the article online for updates and enhancements.

This content was downloaded from IP address [Link] on 28/01/2021 at 02:06

ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

Modeling Multivariate Time Series by Vector Error

Correction Models (VECM) (Study: PT Kalbe Farma Tbk.
and PT Kimia Farma (Persero) Tbk)

L Loves1,3, M Usman1, Warsono1, Widiarti1, E Russel2

1
Department of Mathematics, Faculty of Mathematics and Natural Sciences, Universitas
Lampung, Indonesia
2
Department of Management, Faculty of Economic and Business, Universitas Lampung,
Indonesia
3
Graduate Students of Mathematics, Faculty of Mathematics and Natural
Sciences,Universitas Lampung, Indonesia

email: luvitaloves@gmail.com1

Abstract. Time series analysis (time series) is one method with the aim to find out events
that will occur in the future based on data and past circumstances. Time series are widely used
in economics, business, environmental science, and finance. The analytical tool that is widely
used to answer quantitative research problems is the Autoregressive Vector (VAR). The VAR
model is used if the data is stationary. If the variable has cointegration and stationary at the
first difference value, the VAR model is modified to become the Error Correction Model
(VECM). Then we can find out the influence of variables with other variables by looking at
the Impulse Response Function and Granger Causality. In this research, PT Kalbe Farma Tbk's
stock data will be analyzed. (KLBF) and PT Kimia Farma (Persero) Tbk (KAEF). The data
used are weekly data from January 2010 to June 2020. Based on data analysis, it is known that
the data is not stationary and there are unit roots. Furthermore, first differencing is done to
make the data stationary. Because there was cointegration, a VECM analysis was performed
and a VECM (p) was obtained with a lag of p = 4. So the best model for this research is
VECM (4) with rank = 2. Causal relationships between variables using Granger Causality
showed that KLBF influenced KAEF in the past. Based on IRF analysis, each variable gives a
fluctuating response with itself and with other variables.

Keywords: VAR model,VECM, cointegration, Granger Causality, Impulse Response

Function

1. Introduction
Time series analysis is one method with the aim to find out events that will occur in the future based
on data and past circumstances. In general, the time series econometrics model is a structural model
because it is based on existing economic theories. In 1980 Christopher A. Sims introduced the VAR
model as an alternative in macroeconomic analysis. The analytical tool commonly used to answer
quantitative research problems is the Autoregressive Vector (VAR). The VAR model is used to
explain the simultaneous variables that have influence on each other. The VAR model is used if the
data is stationary at the level. The data is not stationary at the level but stationary at the first difference
value we will use the Autoregressive Vector in Difference (VARD) if all variables do not have

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

cointegration. When the variables have cointegration and stationary at the first difference value, then
the Error Correction Model (VECM) is used. In this applied statistical research various cases of
multivariate time series data will be examined. Modeling that will be used for multivariate time series
data is the Error Correction Model (VECM) Vector, which will then be seen the causal relationship
between time series variables using Granger Causality, to see the effect of the shock of a variable
against other variables will be used Impulse Response Function (IRF) ). Discussed the relationship
and forecasting between the price indexes of two oil companies in Indonesia using VAR [1].
In this research, modeling will be carried out on the stock data of PT Kalbe Farma Tbk. (KLBF)
and PT Kimia Farma (Persero) Tbk (KAEF). PT Kalbe Farma Tbk., with its subsidiaries, develop,
manufacture, and trade pharmaceutical products in Indonesia. It operates in four segments:
Prescription Pharmacy, Consumer Health, Nutrition, and Distribution and Logistics. The company
was founded in 1966 and based in Jakarta, Indonesia. PT Kalbe Farma Tbk. is a subsidiary of PT Gira
Sole Prima. PT Kimia Farma (Persero) Tbk manufactures and sells medicines, herbal medicines,
iodine, salt, quinine and its derivative products, and vegetable oils in Indonesia, throughout Asia,
Europe, Australia, Africa and New Zealand. The company operates through the manufacturing,
distribution, retail and other services segments. The company was founded in 1817 and based in
Jakarta, Indonesia. Both of these variable data are time series data. So that time series analysis can be
done to make multivariate modeling that can be used for the future. The objectives of this study are
(1) Formulating a Multivariate Time series data model with the Vector Error Correction Model
(VECM) approach. (2) Review the behavior of Multivariate data with Granger Causality. (3) Assess
how the behavior of one variable with respect to other variables in the event of shock and how long
the equilibrium will occur.

2. Literature Review
2.1 Test Cointegration
The concept of cointegration was introduced by Engle and Granger and the development of practical
and inferential estimation methods was given by Johansen. In much of the literature, the time series
Xt is said to be integrated with the sequence process 1, I (1), if (1-B) Xt is stationary and cannot be
reversed. If the time series data is stationary and can be reversed, it says process I (0). In general,
univariate time series Xt is process I(d), if (1-B)dXt stationary and non invertable [2],[3],[4]. Burke
and Hunter proposed the procedure of Johansen’s for estimation and inferencial [5]. If there is
cointegration between variables, then we must test the cointegration ranking. Some cointegration rank
testing methods are as follows: Trace test and Test the maximum eigenvalue. In the cointegration test
Johansen cointegration test is used as follows: It is known that the model 𝑉𝐴𝑅 (𝑝) is
𝑦𝑡 = 𝐴𝑡 𝑦𝑡−1 + ⋯ + 𝐴𝑝 𝑦𝑡−𝑝 + 𝐵𝑥𝑡 + ε𝑡
where y_t is a vector with 𝑘 non stationary variable I (1), 𝑥𝑡 is a vector with 𝑑 deterministic variable,
𝜀𝑡 is an error vector. The equation 𝑉𝐴𝑅 (𝑝) can also be written as
𝑝−1

∆𝑦𝑡 = Π𝑦𝑡−1 + ∑ Γ𝑖 ∆y𝑡−1 + 𝐵𝑥𝑡 + ε𝑡

𝑖=1
where
𝑝 𝑝

Π = ∑ 𝐴𝑖 + 𝐼 , Γ𝑖 = − ∑ 𝐴𝑗
𝑖=1 𝑖=1
For testing the hypotheses can be used the trace test :
𝑘

𝐿𝑅𝑡𝑟 (r|k) = −T ∑ log (1 − λ𝑖 )

𝑖=𝑟+1
And statistical test for maximum eigen value
LRmax(r|r + 1) = −T log(1 − λ𝑟+1 )
= 𝐿𝑅𝑡𝑟 (r|k) − 𝐿𝑅𝑡𝑟 (r + 1|k)
for 𝑟 = 0, 1, …, 𝑘 − 1, with the null hypotheses is 𝐻0 ∶ there is 𝑟 cointegration equation.
At the significance level (1 - 𝛼) 100%, 𝐻0 is accepted if the trace test statistic and the maximum
eigenvalue are smaller than the critical value when 𝛼, or 𝑝-value is greater than the significance value

2
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

𝛼 [6]. If there is cointegration between variables, the representation the error-correction VAR model
was modified, so the model became a VECM model [7] [8].

2.2 Vector Autoregressive (Var)

Vector Autoregressive (VAR) is a special form of simultaneous equation system. The VAR model can
be applied if all variables used are stationary, but if the variables in the Y_t vector are not stationary
then the model used is the Vector Error Correction Model (VECM) provided that there is one or more
cointegration relationships between the variables. VECM is a limited VAR that is designed to be used
in non-stationary data which is known to have a cointegration relationship [9].
𝑦𝑡 = 𝐴1 𝑦𝑡−1 + ⋯ + 𝐴𝑝 𝑦𝑝−1 + 𝜀𝑡
where,
𝑦𝑡 : is vector of observation,
𝐴 : matrix of parameter,
𝜀𝑡 : vektor error
If the data used is stationary at the same differencing level and there is cointegration, then the
VAR model will be combined with the error correction model into the Vector Error Correction Model
(VECM) [7].

2.3 Vector Error Correction Model (VECM)

VECM is a limited VAR model designed to be used in non-stationary time series but has a
cointegration relationship between variables. VECM is very useful because it can estimate the short-
term effects between variables and the long-term effects of time series data. The general form of
VECM (p) where p is the lag of endogenous variables with cointegration rank r ≤ k is as follows [2]:
𝑝−1
∆𝑦𝑡 = Π𝑦𝑡−1 + ∑𝑖=1 Γ𝑖 ∆𝑦𝑡−𝑖 + 𝐷𝑡 + ε𝑡
where:
∆ = operator differencing, where ∆𝑦𝑡 = 𝑦𝑡 − 𝑦𝑡−1,
𝑦𝑡−1 = vector variable endogenous with lag 1,
εt = kx1 vector residuals,
𝐷𝑡 = kx1 vektor constant,
Π = matrix coefficient of cointegration (Π = 𝛼𝛽t ; 𝛼 = vector adjustment, kxr matrix and 𝛽 =
matrix cointegration (long-run parameter) (k × r))
Γ𝑖 = kxk matrix coefficient the ith variable endogenous.

2.4 Test For Normality Of Residuals

Residual normality test is used to determine the residual normality in a multivariate model. The
normality test is carried out using the Jarque-Bera (JB) Test of Normality. This test uses a measure of
skewness and kurtosis. Jarque-Bera (JB) used in the normality test on the residual model where the
calculation is done by adding indicators of the number of independent variables or predictors, JB
calculation is as follows:
𝑁 𝑁
𝐽𝐵 = [ 6 𝑏12 + 24 (𝑏2 − 3)2 ]
Where:
𝑁 = number of sample size,
𝑏1 = 𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝑆𝑘𝑒𝑤𝑛𝑒𝑠𝑠
𝑏2 = 𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝐸𝑥𝑐𝑒𝑠𝑠 𝐾𝑢𝑟𝑡𝑜𝑠𝑖𝑠
where Jarque-Bera (JB) Test of Normality with chi-square 𝑥 2 distribution with degrees of freedom 2
[10].

2.5 Granger Causality

Granger causality is used to see short-term relationships in the form of reciprocity between variables
in a vector. A stable VAR is defined as follows:
𝑦1𝑡 𝐴11,1 𝐴12,1 𝑦1𝑡−1 𝐴11,𝑝 𝐴12,𝑝 𝑦1𝑡−𝑝 𝑒1𝑡
𝑦𝑡 = [𝑦 ] = [ ] [𝑦 ]+ ⋯+ [ ][ ] + [𝑒 ]
2𝑡 𝐴21,1 𝐴22,1 2𝑡−1 𝐴21,𝑝 𝐴22,𝑝 𝑦2𝑡−𝑝 2𝑡

3
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

𝑦𝑡 is consist of vector 𝑦1𝑡 and 𝑦2𝑡 . 𝑦2𝑡 is not grager causality for 𝑦1𝑡 if coefficient matrix of
parameter VAR namely 𝐴21,𝑖 = 0 for i=1,2,..., p[2].

The Granger Causality Test is based on the F test which attempts to determine if there is a change
in one variable due to a change in another variable. A variable X is said to be a Granger Cause
variable Y, if the previous value of X can predict the current Y value.
VAR Model:
𝑝
𝑦𝑡 = ∑𝑖=1 ∅𝑖 𝑦𝑡−𝑖 + 𝜀𝑡

If all the coefficients ∅ on the lag value of 𝑦 are significant then X Granger Causal Y. If X Granger
Causal Y and not vice versa, it is called indirect causality. If causality is found in both, from X to Y
and from Y to X, then it is called bidirectional causality [11] [12] [13].

2.6 Impulse Response Function (IRF)

The Impulse Response Function is a method used to see the response of an endogenous variable to
shock given by another variable. A Vector Autoregressive (VAR) can be written in the form of a
Vector Moving Average (VMA) that allows us to see various responses from veriable in the VAR
system. The VAR model can be written in the MA vector (∞) as

And the matrix has an interpretation as follows:

Row i, column j element identifies the consequence of increasing one unit in the innovation of
variable j on the date t (μjt) for the value of the i variable at time t + s (Xi, t + s), holding all other
innovations at all constant dates. If the first μt element is changed by δ1, at the same time, the second
element is changed by δ2, .., and element n by δn, then the combined effect of this change on the
vector value Xt + s will

Plot of row I, column j element of Ψs is called Impulse Response Function (IRF).

3. Results And Discussion

The first step that must be passed to get the VECM estimate is to test the stationarity of each variable's
data. Stationary data is needed to influence the results of the VECM estimation test. In this study, to
detect whether or not the stationary variable of each variable data, it can be seen with a time series
plot, ACF graph (Autocorrelation Function), and Augmented Dickey-Fuller Unit Root Test.

Figure 1. Trend and correlation analysis for KLBF and for KAEF

Table 1. Augmented Dickey Fuller Unit Roots Test

Variable Type lags ρ Pr < ρ Tau Pr <Tau
Kurs KLBF Zero mean 3 -0.4354 0.5837 -0.68 0.4204

4
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

Single mean 3 -19.7479 0.0113 -3.20 0.0215

trends 3 -21.6944 0.0454 -3.26 0.0747
Kurs KAEF Zero mean 3 -0.6201 0.5441 -0.58 0.4652
Single mean 3 -3.7952 0.5605 -1.32 0.6227
trends 3 -2.7430 0.9466 -0.85 0.9589

From Figure 1, the time series plot shows that the two variables above are not stationary because
they still contain elements of trend. Furthermore, the instability of the data is also shown by the ACF
graph where from lag 1 to the next lag falls slowly linearly near zero, this shows that the coefficient of
autocorrelation is significantly different from zero. From table 1, all variables contain unit roots or are
not stationary at the level. This can be seen in the p value of the statistical value Tau (τ) all types of
testing for each variable is greater than the significance limit used, namely α = 0.05, so the data is not
stationary (there is a unit root). Thus it can be said that all the variables above contain unit roots or are
not stationary. Since all veriabels are not stationary at the level level, the first differencing is
performed on the data, then checked again using time series plots, ACF charts and unit root tests.

Figure 2. Trend and correlation analysis for KLBF and for KAEF

Table 2. Augmented Dickey Fuller Unit Roots Test

Variable Type lags ρ Pr < ρ Tau Pr <Tau
Kurs KLBF Zero mean 3 -462.644 0.0001 -11.24 <.0001
Single mean 3 -464.561 0.0001 -11.23 <.0001
trends 3 -466.902 0.0001 -11.23 <.0001
Kurs KAEF Zero mean 3 -305.689 0.0001 -9.91 <.0001
Single mean 3 -305.725 0.0001 -9.89 <.0001
trends 3 -314.081 0.0001 -9.97 <.0001

From Figure 2 in the time series plot it can be seen that the two variables are stationary to the mean
and variance because they no longer contain an element of trend. Furthermore, the stationarity of the
data is also shown by the ACF graph where from lag 0 to the next lag it slowly decreases
exponentially to zero. So it can be concluded that the five variables above are stationary to the mean
and variance. From Table 2, all variables no longer contain unit or stationary roots in the 1st
Differencing. This can be seen in the p value of statistical Tau all types of testing for each variable is
smaller than the significance limit used, α = 0.05, so that the data is stationary (there is no unit root).
Thus it can be said that all of the above variables do not contain unit roots or stationary data.

3. 1 Test for Lag Optimal

VECM estimates are very sensitive to the lag length of the data used. The length of the lag is used to
determine the time needed for the effect of each variable on its past variable. In this study,
determining the length of the lag is done by looking at the smallest value of the information criteria.
Determination of the optimum lag as follows:

Table 3. Lag Optimal

Information criterion VAR(1) VAR(2) VAR(3) VAR(4) VAR(5)
AICC 17.36952 17.36942 17.3686 17.35107* 17.36112

5
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

HQC 17.38788* 17.39994 17.41118 17.40565 17.42762

AIC 17.36939 17.36908 17.36793 17.34995* 17.35944
SBC 17.41668* 17.448 17.47856 17.4924 17.53379
FPEC 34948784 34938002 34897684 34276010* 34603014

Based on table 3, it can be seen that the optimal lag length lies in lag 4. The selection of lag 2 as
the optimal lag is based on the smallest values of AICC, AIC, and FPEC. So cointegration testing will
be carried out in lag 4.

3.2 Test Cointegration

Cointegration testing is used to determine the long-term relationship of each variable. The
requirement in estimating VECM is that there is a cointegration relationship in it. If there is no
cointegration relationship, then the VECM estimate is canceled, but must use the VAR (Vector
Atouregression) model. The cointegration test used in this study is the Johansen cointegration test.

Table 4. Table cointegration

H0: Rank=r H1: Rank>r Eigenvalue Trace Pr > Trace
0 0 0.1979 227.9709 <.0001
1 1 0.1807 108.2325 <.0001

ased on Table 4, it can be seen that the p value for rank = 1 is smaller than the significance limit
used, namely α = 0.05, so there is not enough evidence to reject H1: rank> r. Thus it can be said that
there is a cointegration relationship between variables with rank = 2. Because the data used there is a
cointegration relationship, the VAR (p) model used is VECM (p) with rank = 2.

3.3 Selection of VECM(p)

Selection of VECM(p) based on the information criterion of AICC, HQC, AIC, SBC and FPEC, the
best VECM(p) is as follows:

Table 5. Selection VECM(p)

Information criterion VECM(1) VECM(2) VECM(3) VECM(4) VECM(5)
AICC 17.36952 17.36942 17.3686 17.35107* 17.36112
HQC 17.38788* 17.39994 17.41118 17.40565 17.42762
AIC 17.36939 17.36908 17.36793 17.34995* 17.35944
SBC 17.41668* 17.448 17.47856 17.4924 17.53379
FPEC 34948784 34938002 34897684 34276010* 34603014

Based on Table 5 it can be seen that the smallest values of AICC, AIC, and FPEC are found in VECM
(4). So that VECM (4) was chosen as the best model.

3.4 The estimation Parameter of VECM(4) with rank r=2

Based on the above analysis, VECM (4) was selected as the best model with rank r = 2. Next, we will
estimate the model for VECM (4) as follows:

Table 6. Long-Run Parameter Beta

Estimates when RANK=2
Variable 1 2
KLBF 0.04208 0.01324
KAEF -0.00844 0.01620

6
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

Table 7. Adjustment Coefficient Alpha

Estimates When RANK=2
Variable 1 2
KLBF -23.09162 -10.51971
KAEF 14.39345 -50.96496

Table 8. Parameter Alpha *

Beta' Estimates
Variable KLBF KAEF
KLBF -1.11094 0.02450
KAEF -0.06902 -0.94739

Table 9. Model Parameter Estimates

Equation Parameter Estimate Standard Error t Value Pr > |t| Variable

D_KLBF AR1_1_1 -1.11094 0.09739 KLBF(t-1)

AR1_1_2 0.02450 0.04034 KAEF(t-1)
AR2_1_1 0.00790 0.08104 0.10 0.9224 D_KLBF(t-1)
AR2_1_2 0.00962 0.03526 0.27 0.7851 D_KAEF(t-1)
AR3_1_1 -0.02219 0.06415 -0.35 0.7296 D_KLBF(t-2)
AR3_1_2 -0.03383 0.02925 -1.16 0.2479 D_KAEF(t-2)
AR4_1_1 -0.13230 0.04263 -3.10 0.0020 D_KLBF(t-3)
AR4_1_2 -0.05164 0.02018 -2.56 0.0108 D_KAEF(t-3)
D_KAEF AR1_2_1 -0.06902 0.21217 KLBF(t-1)
AR1_2_2 -0.94739 0.08788 KAEF(t-1)
AR2_2_1 0.11690 0.17653 0.66 0.5081 D_KLBF(t-1)
AR2_2_2 -0.08981 0.07681 -1.17 0.2428 D_KAEF(t-1)
AR3_2_1 -0.02170 0.13974 -0.16 0.8767 D_KLBF(t-2)
AR3_2_2 -0.04961 0.06371 -0.78 0.4365 D_KAEF(t-2)
AR4_2_1 -0.00970 0.09288 -0.10 0.9168 D_KLBF(t-3)
AR4_2_2 -0.04136 0.04396 -0.94 0.3472 D_KAEF(t-3)

Based on the parameter estimation results, the VECM estimation (4) is obtained, i.e.

∆𝑌𝑡 = Π𝑌𝑡−1 + Γ1 ∆𝑌𝑡−1 +Γ2 ∆𝑌𝑡−2 + Γ1 ∆𝑌𝑡−3 + 𝜀𝑡

−1.11094 0.02450 0.00790 0.00962

∆𝑌𝑡 = [ ]𝑌 + [ ] ∆𝑌𝑡−1
−0.06902 −0.94739 𝑡−1 0.11690 −0.08981
−0.02219 −0.03383
+[ ] ∆𝑌𝑡−2
−0.02170 −0.04961

7
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

−0.13230 −0.05164 𝜀𝑡1

+[ ] ∆𝑌𝑡−3 + [𝜀 ]
−0.00970 −0.04136 𝑡2

Table 10. Schematic Representation of Cross Correlations

of Residuals
Variable/Lag 0 1 2 3 4 5 6 7 8 9 10 11 12
KLBF ++ .. .. .. .. .. .. .. .. .. .. .. ..
KAEF ++ .. .. .. .. .. .. .. .- .+ .. .. ..
+ is > 2*std error, - is < -2*std error, . is between

Table 11. Portmanteau Test for Cross

Correlations of Residuals
Up To Lag DF Chi-Square Pr > ChiSq
5 4 5.86 0.2099
6 8 8.06 0.4272
7 12 9.46 0.6636
8 16 17.40 0.3604
9 20 24.82 0.2082
10 24 25.74 0.3663
11 28 26.92 0.5227
12 32 30.73 0.5308

3.5 Normality Residual

Figure 3. Prediction Error Normality for KLBF and for KAEF

Table 12. Univariate Model White Noise Diagnostics

Variable Durbin Normality ARCH
Watson
Chi-Square Pr > ChiSq F Value Pr > F
KLBF 1.98909 241.34 <.0001 4.57 0.0330
KAEF 2.00446 2234.35 <.0001 41.99 <.0001

8
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

Table 13. Univariate Model AR Diagnostics

Variable AR1 AR2 AR3 AR4
F Value Pr > F F Value Pr > F F Value Pr > F F Value Pr > F
KLBF 0.02 0.8993 0.01 0.9877 0.03 0.9941 0.03 0.9986
KAEF 0.00 0.9566 0.00 0.9976 0.00 0.9999 0.00 1.0000

Tables 12 and 13 are used to examine residual white noise on the condition of a univariate
equation. The table shows a statistical test for distribution normality using the Jarque Bera normality
test. Table 12 shows the p-values for KLBF and KAEF <0.05, meaning that the residuals are normally
distributed. From Figure 3 it can be seen that the residual is approaching the normality line.

3.6 Test for Stability Model

The model stability test is used to see whether the model is stable or not.

Table 14. Roots of AR Characteristic Polynomial

Index Real Imaginary Modulus Radian Degree
1 0.48209 0.03817 0.4836 0.0790 4.5274
2 0.48209 -0.03817 0.4836 -0.0790 -4.5274
3 0.04416 0.65159 0.6531 1.5031 86.1226
4 0.04416 -0.65159 0.6531 -1.5031 -86.1226
5 -0.00620 0.38205 0.3821 1.5870 90.9301
6 -0.00620 -0.38205 0.3821 -1.5870 -90.9301
7 -0.50635 0.00000 0.5064 3.1416 180.0000
8 -0.67399 0.00000 0.6740 3.1416 180.0000

Based on Table 14, you can see the modulus value <1. So that VECM (4) is a model that is feasible to
use.

3.7 Test for fit the Model

The model fit test can be seen from the ANOVA table of the univariate model to determine the
significance of the model. Based on the equation of the VECM model (4) written univariately, the
model feasibility test is as follows:

Table 15. Univariate Model ANOVA Diagnostics

Variable R-Square Standard Deviation F Value Pr > F
KLBF 0.5688 48.41149 32.32 <.0001
KAEF 0.5423 18.57153 29.03 <.0001

Based on Tabel 15, the univariate F-test are 32.32 and 29.03 with p-values <0.0001 for both KLBF
and KAEF respectively.

9
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

3.8 Analisys of Granger-Causality

Granger-Causality Test is intended to determine the causal relationship of each independent variable
on the dependent variable. The Granger-Causality test is based on the wald-test with the chi-square
distribution or F-test. The null hypothesis in the Granger-Causality test is where group one is
influenced by itself not by group two.
Table 16. Granger Causality Wald Test
Test Group Variables Pr > ChiSq Conclusion
1 Group 1 Variables : KLBF 0.0056 Reject H0
Group 2 Variables : KAEF
2 Group 1 Variables : KAEF 0.5926 Not enough evidence to reject H0
Group 2 Variables : KLBF

Based on table 16, in test 1, the p-value <0.05 reject H0 means that KLBF is affected by KAEF. In
test 2, p-value> 0.05 was obtained, it means that there was not enough evidence to reject H0. So in the
second test, KAEF is affected only on itself and not on KLBF.

3.9 Impulse Response Function (IRF)

Figure 4. Response to Impulse in KLBF

Based on Figure 4, if the Impulse Response Function (IRF) graph experiences a shock of one
standard deviation it will affect the KAEF variable and itself. If the IRF chart approaches the point of
equilibrium or returns to the zero line, it means that the response of the variable to show other
variables is getting lost so that the shock does not leave a permanent effect on the variable. Shock one
standard deviation at ITMA, because ITMA gives a fluctuating response from the first week to the
ninth week. In the first week to the second week the response is negative. The third and fourth weeks
provide positive responses. The fifth and sixth week gives positive responses. The seventh and eighth
week gives positive responses. The ninth week onwards the response begins to approach the point of
balance and positive response. Shock one standard deviation at ITMA, because ELSA gives a positive
fluctuating response from the first week to the seventh week. In the first week and second week
negative responses. In the third week and the fourth week the value dropped but the response was
positive. In the fifth week and the sixth week the response is negative. Then in the seventh week it
starts to strike a balance point.

10
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

Figure 5. Response to Impulse in KAEF

Based on Figure 5, if the Impulse Response Function (IRF) graph experiences a shock of one
standard deviation it will affect the KAEF variable and itself. If the IRF chart approaches the point of
equilibrium or returns to the zero line, it means that the response of the variable to show other
variables is getting lost so that the shock does not leave a permanent effect on the variable. Shock one
standard deviation at KAEF, because KLBF gives a fluctuating response from the first week to the
third week. Then in the fourth week onwards it does not fluctuate and gives a positive response
because it is above point 0. Shock one standard deviation at KLBF, because KAEF gives a fluctuating
response from the first week to the third week. Then in the fourth week and so on it does not fluctuate
but gives a negative response because it is below the 0 point.

4. Conclusion
Based on the analysis of KLBF and KAEF time series data per week during January 2010-June 2020.
This study examines the relationship between KLBF and KAEF, there is a cointegration relationship
between KLBF and KAEF stock data with rank = 2. Based on the cointegration test and the smallest
value of the information criteria, the best model is VECM (p) with lag p = 2. Meanwhile, the granger
causality test explains that in test one a p-value of <0.05 starting with H0 means that KLBF is affected
by KAEF. Whereas in the second test, there was not enough evidence to reject H0, meaning that
KAEF stock data was affected only by itself and not by KLBF stock data. Based on IRF analysis,
each variable gives a fluctuating response with itself and with other variables.

Acknowledgements
The authors thanks [Link] and the Indonesia Stock Exchange (IDX), which has provided
data in this study. This study is funded by the Directorate of Research and Community Service,
Deputy Research and Development Ministry of Research and Technology Republic of Indonesia /
National Research and Innovation Agency in accordance with the Research Contract No:
044/SP2H/LT/DRPM/2020. The authors thank the Directorate of Research and Community Service
Deputy Research and Development Ministry of Research and Technology Republic of Indonesia.

References

[1] Warsono at all 2019 Vector Autoregressive With Exogenus Variable Model and Its Application
in Modelling and Forecasting Energy Data : Case Study Of PTBA and HRUM Energy
International Journal Of Energy Economics and Policy 9(2) p 390-398
[2] Lutkepohl H 2005 New Introduction to Multiple Time Series Analysis (Berlin: Springer
Verlaag)
[3] Tsay R S 2014 Multivariate Time Series Analysis: With R and Financial Applications
(Hoboken
New Jersey: John Wiley and Sons, Inc)
[4] Tsay R S 2005 Analysis of Financial Time series 2nd ed (Hoboken New Jersey: John Wiley and
Sons, Inc)
[5] BurkeS.P and Hunter J 2005 Modelling Non-stationary Time Series: A Multivariate Approach
(New York: Palgrave Macmillan)

11
ICASMI 2020 IOP Publishing
Journal of Physics: Conference Series 1751 (2021) 012013 doi:10.1088/1742-6596/1751/1/012013

[6] Rosadi D 2012 Ekonometrika dan Analisis Runtun Waktu Terapan dengan Eviews
(Yogyakarta: Penerbit Andi Offset)
[7] Asteriou D and Hall S G 2007 Applied Econometrics: A Modern Approach (New York:
Palgrave
Macmillan)
[8] Wei W W S 2019 Multivariate Time Series Analysis and Applications (Hoboken New Jersey:
John Wiley and Sons Inc)
[9] Enders W 2015 Applied Econometric Time Series (USA: John Wiley and Sons Interscience
Publication)
[10] Jarque C M and Bera AK 1987 A test for normality of observations and regression residuals
International Statistical Review 55 p 163-172
[11] Brooks and Chris 2008 Introductory Econometrics for Finance 2nd ed (New York: Cambridge
University Press)
[12] Warsono at all 2019 Modeling and Forecasting by the Vector Autoregressive Moving Average
Model for Export of Coal and Oil Data International Journal of Energy Economics and
Policy 9(4) p 240-247
[13] Warsono at all 2020 Dynamic Modeling Using Vector Error-correction Model Studying the
Relationship among Data Share Price of Energy PGAS Malaysia, AKRA, Indonesia, and
PTT PCL-Thailand International Journal of Energy Economics and Policy 10(2) pp 360-
373

12