Scribd
Upload
121 views

Lecture 8 Application of VAR Model

This document discusses various analyses that can be conducted using vector autoregressive (VAR) models, including impulse response functions, variance decomposition, and Granger causality. Impulse response functions show how variables respond over time to shocks, while variance decomposition determines how much of a variable's forecast error is explained by its own shocks versus those of other variables. Granger causality tests whether including lagged values of one variable helps predict another variable, indicating a causal relationship. The document provides examples of implementing these analyses in EViews software.

Uploaded by

Eason Saint
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
121 views

Lecture 8 Application of VAR Model

This document discusses various analyses that can be conducted using vector autoregressive (VAR) models, including impulse response functions, variance decomposition, and Granger causality. Impulse response functions show how variables respond over time to shocks, while variance decomposition determines how much of a variable's forecast error is explained by its own shocks versus those of other variables. Granger causality tests whether including lagged values of one variable helps predict another variable, indicating a causal relationship. The document provides examples of implementing these analyses in EViews software.

Uploaded by

Eason Saint
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 22

TOPIC 8:

APPLICATION OF VAR
MODEL
By:
Assoc. Prof. Dr. Sallahuddin Hassan

SEEQ5133

Applied Econometrics

INTRODUCTION
2

Some analysis using VAR model:

Impulse response functions


(IRFs)
Variance decomposition
Granger causality

SEEQ5133
ECONOMETRICS

APPLIED

IMPULSE RESPONSE
FUNCTION

Impulse response function (IRF)


shows the effects of shocks on the
adjustment path of the variable.
Examines the response of the
dependent variable to shocks in the
error term or exogenous shock:

nominal and real shock


domestic and external shocks
permanent and transitory shocks
SEEQ5133
ECONOMETRICS

APPLIED

IMPULSE RESPONSE
FUNCTION

Pattern of coefficients are IRFs.


IRFs depict:

how the shock spread up over time.


the response of each variable taken in level
to a 1% shock as well as the confidence
interval.

Eviews implementation:

Select View/Impulse and in impulse


definition tab choose residuals-one std.
deviation
SEEQ5133
ECONOMETRICS

APPLIED

IMPULSE RESPONSE
FUNCTION - VECM

SEEQ5133
APPLIED

IMPULSE RESPONSE
FUNCTION - VECM

SEEQ5133
APPLIED

FORECAST ERROR
VARIANCE DECOMPOSITION

Forecast error variance decomposition


(FEVD) explains the proportion of the
movements in a sequence due to its own
shocks versus shocks to other variable.
FEDV:

enables to determine the most fluctuation


sources of the endogenous variables for the
period of study
permits to measure the part of the anticipated
variance of each endogenous variable explained
by the different shocks for the different horizons.
SEEQ5133
APPLIED

FORECAST ERROR
VARIANCE DECOMPOSITION

Variable that is expected to have any


predictive value for other variables
should be put last.
The percentage of variation depends
on:

Correlation between the residuals of a


variable and the residuals of variable
that appear before it in the ordering.
Correlation among innovation.
SEEQ5133
ECONOMETRICS

APPLIED

FORECAST ERROR
VARIANCE DECOMPOSITION
- VECM

SEEQ5133
APPLIED

10

FORECAST ERROR
VARIANCE DECOMPOSITION
- VECM
Variance Decompos ition of REXP_M:
Perio...

S.E.

REXP_M

RFDI_M

RGDP_M

1
2
3
4
5
6
7
8
9
10

61904059
99067994
1.31E+08
1.59E+08
1.84E+08
2.07E+08
2.28E+08
2.48E+08
2.67E+08
2.87E+08

100.0000
95.98935
95.74035
96.09691
96.56349
96.92545
97.03860
96.80910
96.17334
95.09092

0.000000
2.341486
2.541552
2.180861
1.715001
1.363373
1.267284
1.520893
2.188272
3.310043

0.000000
1.669160
1.718093
1.722232
1.721512
1.711182
1.694112
1.670005
1.638391
1.599041

SEEQ5133
APPLIED

RESULT OF ANALYSIS
11

In the short run, impulse of


innovation or shock to REXP
account for 100 percent
variation of the fluctuation in
REXP (own shock).
Shock to RFDI and RGDP can
cause 0.00 percent in the first
period.
SEEQ5133
ECONOMETRICS

APPLIED

RESULT OF ANALYSIS
12

In the long run, impulse of


innovation or shock to REXP
account for 95.09 percent
variation of the fluctuation in
REXP (own shock).
Shock to RFDI and RGDP can
cause 3.31% and 1.59%.
SEEQ5133
ECONOMETRICS

APPLIED

CAUSALITY
13

Refers to the ability of one


variable to predict (and
therefore cause) the other.
Suppose Yt and Xt affect each
other with distributed lags.
This relationship can be
captured by a VAR model.
SEEQ5133
ECONOMETRICS

APPLIED

CAUSALITY
14

Granger (1969) developed


causality test:
A variable Yt is said to Grangercauses Xt ,
if Xt can be predicted with greater
accuracy by using past values of the
Yt rather than not using such past
values, all other terms remaining
unchanged.
SEEQ5133
ECONOMETRICS

APPLIED

CAUSALITY
15

Yt 1 i X t i jYt j 1t
i 1
N

j 1
M

X t 1 i X t i jYt j 2 t
i 1

j 1

i 1

j 1

Yt 1 i X t i jYt j 1 ECTt 1 1t
N

i 1

j 1

X t 1 i X t i jYt j 2 ECTt 1 2t
SEEQ5133
ECONOMETRICS

APPLIED

16

DIRECTION OF
CAUSALITY

Unidirectional causality from Yt to Xt.


The estimated coefficients on the lagged X in Equation
1 is statistically significant. Variable X (Granger) causes
Y.
The estimated coefficients on the lagged Y in Equation
2 is not statistically significant.

Unidirectional causality from Xt to Yt.

The estimated coefficients on the lagged X in Equation


1 is not statistically significant.
The estimated coefficients on the lagged Y in Equation
2 is statistically significant. Variable Y (Granger) causes
X.

SEEQ5133
ECONOMETRICS

APPLIED

17

DIRECTION OF
CAUSALITY

Bilateral causality of Feedback.

The set of lagged Y and X coefficients


are statistically significant different
from zero in both regression.

Independence.

The set of lagged Y and X coefficients


are not statistically significant
different from zero in both regression.
SEEQ5133
ECONOMETRICS

APPLIED

GRANGER CAUSALITY TEST


18

Two tests:

Granger causality test


Sim causality test

Case two stationary variables


Yt and Xt.

yt 10 11 yt 1 12 xt 1 1t form:
Standard/reduced
xt 20 21 yt 1 22 xt 1 2t

SEEQ5133
ECONOMETRICS

APPLIED

19

GRANGER CAUSALITY
TEST

xdoes not G-cause y if xt 1 do not help

in prediction of y , controlling for all


t
other relevant information available at
t 1.
H 0 : x y( xdoes not G-cause y ) H 0 : 12 0
Single equation tests implemented as
Wald tests (F-statistic or 2 -statistic).

SEEQ5133
ECONOMETRICS

APPLIED

20

GRANGER CAUSALITY
TEST

E-views implementation:

View/Lag Structure/Granger
causality-block exogeneity
tests (in VAR) or
Quick/Group
statistics/Granger causality
test/Series List/OK/Lag
Specification/OK
SEEQ5133
ECONOMETRICS

APPLIED

21

GRANGER CAUSALITY
TEST

SEEQ5133
ECONOMETRICS

APPLIED

22

GRANGER CAUSALITY
TEST

SEEQ5133
ECONOMETRICS

APPLIED

You might also like