Hi.

In reality you have two trivariate models (M2, oil prive and CPI) then (CPI, EXCH and TradeB) right?

If you want to test the relationship (long-run and short-run) between these variables (for each model), then you can use an Error correction Model and cointegration. Start with Var. VECM and then Granger causality tests.

You can also try Var model with impulse response functions to see the reaction of each variable, and then use a vecm/ECM for your second model.

Hope this help

Helmi

Also Autoregressive Distributed Lag Models can help you simplify your analysis. There is currently a wider acceptance for ARDL in studies where Helmi suggested. I also agree with him in general. For ARDL, you can see Pesaran et al research with Microfit being the one option used for ARDL estimations. People use Eviews but I am not sure if they do all the regressions manually or have a PROC. Anyway, I recommend ARDL, Bounds Testing approach, and Back Helmi suggestion.

Thank you very much Dr. Helmi and Anees. Much appreciated on your advice. Actually, I have tried VECM in eviews but was unsure of how to go about selecting appropriate lag lengths. and yes, since my CPI is at index of 2005 prices, should I also make oil price as a index of 2005 prices. Currently it is at US$ current oil prices.

Dr. Helmi, are you suggesting me to run two separate equations or it can be done all in one.

It depends on your choice of modelling and relevant Econometric issues. You can but you should be sure what you do is best in theory also with practicalities.

Yes, I agree with Anees, your model depends on what you are looking for. If you do a VECM with 6 variables; I fear that the Granger causality test based on short-run, joint test and long-run woukd be a hard task for you, especially if you lag is 3. VECM requests a long data series to get good results and data in general should be indexed to GDP deflator (2005 in your case).

If you are not sur and you find the task very hard, you can try a simple OLS regression with all your variables, this is the easiest case, again it depends on you.

What I suggest is to find some papers which treat the same issues, this will certainly help you to understand the methodology.

**Do you want to know how to select the optimal lag or you have an issue with the result?

Thank you Dr Helmi and Anees once again. Actually, based on your advise I am thinking to model it like to make Trade balance as a function of real exchange rate, CPI and oil price using VECM. I am also attaching a similar paper which has done something in this line. Do you think the approach is fine is this paper. Regarding the lag, yes I have problem selecting the optimal lag length. Please advise

My function is defined as above, that is trade balance as a function of oil price, cpi and real effective exchange rate. Trade balance(TB) is as a ratio of Exports/Imports, while oil price (oil), REER and CPI is all index at 2005=100. And then I have transformed all variables in its log form. When carrying out unit root tests, for LTB, REER and LCPI, I find that they are non-stationary at their level and stationary at 1st difference either I use none, constant or constant and trend as my specification. But for LOIL, at its level, when I specify none and constant it is not stationary but when I specify constant and trend it is stationary at level. What should I do, so I proceed to 1st difference. At first difference with all its specification it is found to be stationary.

1. As it is, my 3 variable are in its first difference while one is in level, how should I proceed from here ? I know this will create problems. Even if I remove index from oil prices and directly take oil price as US dollar per barrel, same situation is there.

2. I also find the variables to eb having 1 co integrating vector when i tested it using the variables in its level. Was it right thing to do. All in same order.

3. Can i run VECM now and when entering in the model specification in Eviews, should I enter like LTB or dLTB.

Please Advise. Looking forward to all your generous feedback.

- Let's start by the Lag lengh creteria: when you run your VAR, then go to => view=>Lag Structure=> then Lag lengh creteria. Use the max of lag (8 for example, if it does not work, do 6, etc)..you will have a table LR, FPE, AIC, SC and HQ wich indicate the optimal lag. You have to select the best one ..

- you have LTB, REER and LCPI, statitionary at 1st deifference..Ok

- Loil. if it is statuionary at level, it should be stationary at its first difference too.

- You said ' At first difference with all its specification it is found to be stationary" this means that your variables are integrated in order 1.

- For VECM it is like your VAR but with the right lag that you find it in the table as I described above.

However, for Granger causality tes (WALD tests), you have to run an OLS with dLAG and dont forget to add the lag for all your variables and also to calculate your Error Correction Term (ECT).

- Have a look at Eviews Forum, it is full of FAQ and will certainly help.

Best

I can appreciate the details from Helmi.

That should be enough to start with as a novice and I recommend you play around with the Eviews forums for discussions (you need to ask the questions) Just google search your question with terms like Eviews or Eviews forum.

You should then find earlier discussions between groups which are really amazing to read and know new things.

You be Happy with Eviews.

Best

Anees

Thanks Helmi and Anees very much. Appreciated.

1. for lag selection, i have a total observation of 45 and the max it allows to be entered as lags to include is 9 and I do note that the results with (*), the corresponding value is the optimal lag structure. However, as I change lags, the AIC criterai chosen lag also changes(like for 4 chosen, it gives 4 lags, 6 chosen, it gives 6 lags and so on), however the SC criteria gives lag as 1 for any entered lags to include from 1 to 9. So based on this what should I include as my optimal lag. Can I choose 1 as my optimal lag.

2. As for oil price, yes it is stationary at level while others at 1st difference. My confusion is that if it is already stationary at level, will taking first 1st difference make any problem in the regression result.

3. Please see attached word file for my result. Does it look fine and how best to interpret it.

4. When doing Granger Causality tests, variables entered should be difference form or in level meaning [d(ltb) or only ltb) and how do we say that one granger causes another.

Please Advise.

Thanks in advance.

Thanks Mitsis for your feedback. Will definitely consider this.

I have run my model based on some of the advise given above. Please help me in understanding this and let me know if I am on the correct path. I have set out my specific clarification needed in the attachment.

Thanks in advance.

I have run my model based on some of the advise given above. Please help me in understanding this and let me know if I am on the correct path. I have set out my specific clarification needed in the attachment.

Thanks in advance.

Thank you Dr. Pandelis for your wonderful, great and clear explanation. It is very much appreciated and I am certain that like me other upcoming students studying econometrics would also find this comment very useful. I am just being blessed. Thank you once again. I will re-do my model and be back on research gate with revised version.

Some of my feedback are as follows before I revise my estimation:

1. I am very new to econometrics but take keen interest to develop my understanding in this field, thus I am tapping in all the resources I could and your advise is just great. I would also appreciate, if someone can also direct me to some of the research papers which clearly outlines, the basic econometric concepts being used.

2. I did note that my oil price is stationary in first difference, however, I thought that it could pose issues when you are estimating those variables when the order of integration is different. Therefore, I took of significance level as at 1% which made oil price non-stationary in level and stationary in first difference as other variables.

3. Since, you stated that AIC favours models with larger samples, this means that I should use SIC selection method when doing unit root test as well. I had used AIC method in the above estimation and I suppose that the unit root test results would change. On the same note, how should you decide whether to use intercept, trend and intercept and none when doing unit root test. Will it make any issue if some variables are stationary at level for intercept only while some are stationary in level at trend and intercept or at none.

4. Thank you for providing me understanding on the purpose of lag length. Just one thing, if the maximum suggested is 4, when do you actually dont use 4 and try values less than this.

5. Yes, I entered the variables in levels when doing cointegration tests since I read somewhere that for johansens cointegration tests, variables should be entered in its level. But, what buzzes me is that sine you will be entering variables in level in johansen cointegration test and VAR estimation, then what is the puropose of unit root test then, when you find some of your variable stationary in level and some in 1st difference.

6. when interpreting the results of VECM for LR relationship, do we change the variable sign when reporting result, meaning when writing it as a subject of trade balance (ltb). As you SR, the emphasis to be only on significant values based on t-stat of 1.96?

7. As for the issue of including one variable twice, I have only included in my estimation oil price as a index of 2005, that is I have made oil price in 2005 to be 100. However, I just reported in this case the current US oil price per barrel in unit root test to show either using as a index or current price, results are same.

8. I will certainly work on your last comment to revise my model and I have transformed my variables on log form to make interpretation as elasticities. i hope this is fine.

Thank You !!!

Thank you very much Pandelis. Your comments and advise have been very generous and of great help. I have learnt alot. On this note, I shall be as mentioned earlier, revising my model and to see how it has changed. Once again thank you very much and if time does permit you, I shall be most grateful for your any future advise.

I want to thank everyone for the king help they have provided me till to date especially to Mr. Mitsis. I have revised my model based on the comments received above and need some third eye to verify what I have done and where I still need to work on.

Looking forward to your contributions. Thank You.

Looking forward to some helpful comments on my attempt using VECM attached above. Still waiting for some responses.

Thanking in advance.

Dear Prakash

I think you have a set of variables that are interrelated, why you don't try to do with SEM, or simultaneous Equation modelling. It has more theoretic conceivability.

I think that the majority of studies have used the ARDL approach. We get better results using it.

Thank you very much Ahmad, Mohamed and of course Mitsis for the much additional inputs and clarification. Mitisis, a kind request if you possibly have a look at what I have done. I have tried following your approach and have done it accordingly.

Looking forward.

Thanks

All the comments posted here are useful.

Also, there is another source at: http://sayedhossain.com/

Followers of this thread may like to explore there. I found it excellent. 

When a country's exchange rate increases relative to another country's, the price of its goods and services increases. Imports become cheaper. Ultimately, this can decrease that country's exports and increase imports.