Asymmetric Co-integration using Eviews

TIME SERIES ECONOMETRICS WORKSHOP:

"Asymmetric Co-integration, NARDL and Structural VAR"

by Professor Mansor Ibrahim

Session 1: Asymmetric Co-integration using Eviews

• Here is the link to dataset folder: LINK
• Install Eviews Add-in called 'tarcoint' for Threshold Adjustment Model: Eviews Menu --> Add-ins --> Download Add-ins --> tarcoint --> Install).
• Import data into Eviews (Food Price Inflation.xlsx) [or open the file called asymmetric cointegration.wf1]
• Transform original data into natural log:

GENR LFP=log(fp)

GENR LWTI=log(wti)
GENR LY=log(rgdp)

• Open Threshold Adjustment Model from Add-ins Menu.

Option 1: Threshold Adjusted

• Endogenous variables (as separate series or a group): LFP LWTI LY
• Exogenous variables (as separate series or a group): Only if you have
• Number of lags to be used: EMPTY (as default) (Keep empty unless you have strong reason for the certain number of lags)
• Maximum lag order to be tested (if no lags are specified): 6
• Selection criteria for lag length: AIC (as default). Tips: Long data - use AIC, short data - use SIC, since AIC is bias towards the higher order, SIC - towards the lower order)
• Adjustment model: Threshold
• Threshold value (leave blank to initiate the search procedure): 0 (as default)
• The proportion of the sample trimmed at either ends: 0.15 (as default)
• Sample period: 1971 2012
• Monte Carlo experiment for critical values: SELECT
• Number of simulations: 1000 (for this tutorial) (otherwise: 10000; higher - better, more robust)
• Significance level (i.e. alpha): 0.05 (as default)
• Coefficient of lagged changes (i.e. delta): 0.6 (as default)
• Show output table: SELECT (as default)
• OK

Option 1: Threshold Adjusted results:
Note: Threshold value is zero

• OUTPUT: Is there cointegration? 

Null hypothesis: p1=p2=0. 

If failed to reject --> No cointegration. 

If rejected --> At least one of p (p1 or p2) is significant (here the Null hypothesis is p1=p2)

1. Threshold value = 0,  
2. F-equal = 0.152953   <   (1.768753)* (C.V. at 5%), --> failed to reject null (Null: Symmetric)
3. T-max = -3.056021   >   (-2.536955)* (C.V. at 5%), --> reject null (there is long-run relationship/cointegration).
4. F-joint (phi) = 10.188170   >   (7.749866)* (critical value for 5% from Monte Carlo), then reject null hypothesis (there is no long-run relationship/cointegration).

Option 2: Momentum Adjusted

Option 2: Momentum Adjusted results:

Note: Keep threshold value empty

Note: If no cointegration, no point of testing whether symmetric or not.

Let's assume that we find cointegration and asymmetry.

Next, to create variables with positive and negative variations based on MTAR:

• Run OLS regression: LFP c LY LWTI
• Save residuals: Proc --> Make residual series --> RES --> OK
• Create variables for positive and negative thresholds:

GENR DRES=D(RES)

GENR MTAR=DRES(-1)>=0

GENR ZPLUS=MTAR*RES(-1)

GENR ZMINUS=(1-MTAR)*RES(-1)

• Transform data into differenced form:

GENR DFP=D(LFP)

GENR DLY=D(LY)

GENR DWTI=D(LWTI)

• Run OLS regression:

DFP c DFP(-1 TO -3) DY DY(-1 TO -3) DWTI DWTI(-1 TO -3) ZPLUS  ZMINUS

Note: Trim down number of lags until p-values are significant.

DFP c DFP(-1) DY DWTI  ZPLUS  ZMINUS

To create variables with positive and negative variations based on TAR:

GENR TAR=RES(-1)>=0

GENR TPLUS=TAR*RES(-1)

GENR TMINUS=(1-TAR)*RES(-1)

The rest of analysis is similar with MTAR above.

NOTE: To install Go-Global client onto your computer, download the application from here. 
             [Go to GO-Global 5.0 for Windows --> GO-Global Clients --> Choose Windows or MacOS X]