Steps of estimating VECM-causality relationship analysis?

[tsaparyan]

Hello everyone, how do you do?
My name is Ryan and I'm doing my final paper on relationship between CPI and WPI : Causality Analysis. Data in use are monthly CPI and WPI from 1980:1 - 2011:12, which serve as the variables. As many literature suggests, if the series are I(1) and found to be cointegrated, therefore VECM is used. I want to make sure if what I have done is correct. I'm hoping for corrections and suggestions from anyone here in this forum, especially the helpful and attentive admins. It will be very much appreciated. Here are the steps that I've done so far:

1. Checking the stationarity of the data using ADF and PP test, which is found to be stationary in 1st difference forms and not in their level forms. Should the series be checked in index numbers or in log forms? (I went ahead using the index numbers,please do correct me if I'm wrong). RESULT#1 of ADF and PP test are not shown.
2. Then I highlighted both series and "open as VAR", accepting all the default settings, obtaining an initial VAR result. RESULT#2.
3. Next, checking lag length criteria through clicking the "view" toolbar >> lag structure >> lag length criteria accepting the default of including 8 lags and obtain result of " * " at lag 7 in the LR, FPE and AIC column and at lag 2 in SC & HQ column. (I used lag 7 as more of the criterions suggested that.) RESULT#3.
4. Afterwards I check for cointegration through the "view" toolbar, clicking "cointegration test",using assumption 6 for the summary with lag intervals at "1 7".The result shows a " * " on the "quadratic intercept trend" column of AIC (assumption 5). Then click "cointegration test" again using assumption 5 with the results of Trace and Max-eigenvalue test indicates 1 cointegrating eqn(s) at the 0.05 level. RESULT#4.
5. Since the series are cointegrated, I went ahead and click "estimate" on the tool bar, change the VAR type to Vector error correction, endogenous variables: cpi wpi , lag interval: 1 7 and obtain the results. RESULT#5.
6. For stability test, clicking the lag structure >> AR roots and AR Graph, thus obtaining RESULT#6.
7. Finally, it is set for the Granger Causality/Block Exogeneity Tests by clicking the Lag structure in the menu of View toolbar obtaining RESULT#7. with D(CPI) granger cause D(WPI).

From these steps above that I've done, will it be sufficient enough for the estimation and analysis of my paper? If I did further estimates of Impulse Response Function and Variance Decomposition, will it help enrich the analysis? I have attached the EViews results of RESULT#2 - RESULT#7 as references, please do take a look. Thank you so much for your kind help.

[Bandleader]

Hi!

Too sad noone replied. The forum is rich in questions but poor in answers... but anyway,

looks good what you have done, but one thing looks a little... unexplanied.
How did you determine to use option 5 in the cointegration test? Just using the 6 and then checking für significance by looking for ***, is it the right way? I am not very sure about that and if you can explain maybe you are right. But I think it is important to make assumptions about the series you use and the cointegration relation which combines both series. Both series are non-stationary, great! I think if you suggest the cointegration relation between bot series is stationary you should use option 2, what means a constant in the cointegration equation. I both have zero mean you do not need the constant and use option 1. If there is a deterministic trend in the cointegration relation, what means they diverge or converge over time, thus they have a trend-stationary relation, you should use option 4. If you think the cointegration relation has an underlying stochastic trend option 3 is your choice (not sure about that). Option 5, no idea
I think the eviews user's guide II page 687 give an unsufficient explenation of that model specification. You should really take a look into Johansen (1995, p. 80-84) to make sure. I am struggling with the same question on a paper I am writing. So if you have an idea, just post it ;-)

And did you test your series for trend-stationarity using ADF and PP test (maybe you should include the KPSS-test which is a litte more strict and reliable I think)? I don't know what to do with trend-stationary series. can they be tested for cointegration?

Hope you can do your work, maybe you did already :D