Dear Victor,

 

Thank you for your questions.  I’ve read them and I think I am not the best person to answer these.  Your best option is to subscribe to the ALLSTAT mailing list and post these to the list using the address [log in to unmask] .  This is an email list of statisticians around the world and I am sure you would find someone who would be able to respond to your questions free of charge and in a reasonable timeframe.

 

To subscribe to ALLSTAT (which is free), go to https://www.jiscmail.ac.uk/cgi-bin/webadmin?A0=allstat and follow the instructions in the subscribe link.

 

I hope this is helpful

 

Regards

 

Nigel Marriott

Chartered Statistician

 <http://www.marriott-stats.com/> www.marriott-stats.com

 

Ground Floor, 21 Marlborough Buildings, Bath BA1 2LY, United Kingdom

Tel (mobile)  +44 (0)773 4069997

Tel (office)    +44 (0)1225 489033

Fax                  +44 (0)870 6221969

 

Marriott Statistical Consulting Limited, Company No. 5577275, VAT No. 883304029 

Registered in England,   Registered Office – Equity House, 4-6 School Road, Tilehurst, Reading, RG31 5AL

 

From: victor chukwuemeka [mailto:[log in to unmask]] 
Sent: 14 June 2010 16:30
To: [log in to unmask]
Subject: stats questions

 

Dear Nigel,

 

Many thanks for taking some time to look at some question I have below. As I mentioned, I am a Ph.D student in London and I am writing part of a paper which examines how the trade balance responds to the exchange rate. I am using the trade balance as the dependent variable and the exchange rate, domestic income and foreign income as the independent variables. I am confident with the stata commands and the models used but need to know when and why I use them. I use stata primarily and would like to continue to do so.

 

1. I have created log values of all the variables then conduct stationary tests. They are mostly non stationary. When I generate the first differences and find they are stationary to order 1. What lag length do I choose to do stationary tests? Does it matter? What is the significance of the lag length for the stationary tests?

 

2. I then perform cointegration tests by testing the stationarity of the error term of the regression model’s 4 variables. (The Engle-Granger test) I find that most multivariate series are non stationary and therefore not cointegrated)

 

3. I then perform the Johansen test and find that from the r=0 criteria, the trace statistic is lower than the critical value, indicating that I cannot reject this and therefore there is no cointegration.

 

My questions are: 

 

1. Is it true to say that I cannot at all use a VAR model if there is no cointegration and that I must use a Vector Error Correction (VEC) model to correct? 

 

2. After discovering the variables are non-stationary and generating first differences, could I now use the VAR model on the first difference (the first differences are now stationary)? i.e. the command would look something like: var d.tradebal  d.exchangerate  d.domincome  d.forincome, lag(#).  I have been told that the VAR model incorporates the first difference anyway so does it make sense to perform a VAR on the first difference? 

 

3. Also, if I use the VECM, having generated the log of the variables, I have been told to take the first difference as the variables are non stationary, then use the VECM. Most of the Stata manuals use the log of the variables only and not the log of the first differences. Is this true? I am confused about this. Does the VECM calculate the first differences anyway? 

 

4. I understand that the VECM tells us about the short run dynamics even though we use the lags. How do we test the long run in Stata using VECM? Also, the output in Stata when using the VECM gives a D_logvariable. What does the D_ mean? Is this the difference? 

 

5. What is the rule with using the granger causality on VECM. Is it necessary? 

 

6. What happens if I wish to perform a VAR or VECM on 4 variables, one of which is stationary but the other 3 non-stationary? 

 

7. What happens when the dickey fuller cointegration test for the residuals is overwhelmingly stationary with p values of 0.000 but the johansen test cannot reject the r=0 as the trace statistic is lower than the critical value? Could this be because one of the variables is stationary whereas the other 3 are not? Which test do should I have more confidence in for my results?

 

I am actually trying to see how long (how many quarters) it takes for the trade balance to become positive from a depreciation in the exchange rate. i.e. after how many lags. From the Stata manuals, I cannot see how the VECM shows me this as it would in the normal VAR. 

 

8. What would the difference be between using the standard regression command in Stata: reg var1 var2 var3 etc. with the differenced and lagged commands using the VAR var1 var2 var5, lag(#)? 

 

I really hope you could answer these questions. I am confident with the commands but the rules seem to differ from manual to manual and literature to literature. I really hope you can help. 

 

Many thanks and best regards,

 

Victor Chukwuemeka (Ph.D student). 

 


You may leave the list at any time by sending the command

SIGNOFF allstat

to [log in to unmask], leaving the subject line blank.