Structural VAR using Eviews

TIME SERIES ECONOMETRICS WORKSHOP:

"Asymmetric Co-integration, NARDL and Structural VAR"

by Professor Mansor Ibrahim

Session 3: Structural VAR using Eviews

Types of VAR: Reduced Form (approximate) and Structural Form (based on theory)

This technique is strongly based on theoretical relationship between the variables.

1. Import data

2. Transform variables:

GENR LY =LOG(NGDP/GDPDEF*100)

GENR LCPI=LOG(CPI)

GENR LM2=LOG(M2)

GENR LOIL=LOG(BRENT)

GENR LSP=LOG(KLCI)

3. To plot graphs: Menu --> Quick --> Show --> LY LCPI LM2 LOIL LSP --> OK --> 
--> View --> Graph --> Multiple graphs

We find seasonal pattern in LY. What is the solution?

• Option 1. Create dummy variable,
• Option 2. Remove seasonality: Open LY variable --> Plot Graph --> Proc --> Seasonal Adjustment --> Census X12 --> as default (created new variable LY_SA)

4. Identify the number of lags:

Menu --> Quick --> Estimate VAR --> Unrestricted VAR

LYSA LCPI LM2 MMR LSP LOIL

Output:

View --> Lag Structure --> Lag length criteria (8 by default)

Go back to VAR Specification window (click on Estimate button on Menu bar) and specify Lag Intervals for Endogenous as (1 5). 

Output:

The results are based on reduced form.

To interpret this, we need to introduce shocks.

Impulse Response

Go to Menu --> Object --> New Object --> Matrix-Vector-Coef --> matrixB

--> OK 

--> Rows=6 and Columns=6 --> Edit table --> Type "NA" on diagonal --> Turn off Edit Mode --> Close window.

Result:

Similarly, we create MatrixA but with imposed restrictions.
Go to Menu --> Object --> New Object --> Matrix-Vector-Coef --> matrixA

--> OK

--> Rows=6 and Columns=6 --> Edit table --> Fill the cells --> Turn off Edit Mode --> Close.

Go back to VAR output --> Menu --> Proc --> Estimate Structural Factorization --> Matrix --> Short-run pattern --> A=matrixA and B=matrixB

--> OK

Look here at Chi-square and probabilities

Click on Impulse Response button --> Impulse Definition --> Structural Decomposition --> Multiple Graphs --> Analytic --> OK

Interpretation of graphs:

• if the zero line is within the confidence intervals, then not significant.
• if not, then we can explain according to blue line's behaviour.

View -->Variance Decomposition --> Table