Dear all,
I obtained a statistically significant adjustment term (at 1% leve) but the cointegrating evidence is only significant at the 10% level. Is it okay to conclude that that is 'marginal or minimal' cointegration between the dependent variable and the covariates?
Any assistance will do....thanks!
Hi Ngozi,
You have the same "first name" as my younger sister! Yes, there are two ways of concluding cointegration. The first is by using PSS (2001) F test and the second is the use of the t-test on the ecm(-1) term. You can read my paper on this, see 'A convenient method ..... see my RG for correct title. The F-test suggested by PSS (2001) is affected by the number of the first differenced terms in the RhS of the model. Therefore, it is always much better to reinforce your cointegration using the t-ratio. Therefore, you have cointegration!
Arize, A.C. (2017)
"A Convenient Method for the Estimation of ARDL parameters and Test Statistics: reaiUSA trade balance real effective exchange rate relation" in IREF, 50, 75-84
Hi Ngozi,
You have the same "first name" as my younger sister! Yes, there are two ways of concluding cointegration. The first is by using PSS (2001) F test and the second is the use of the t-test on the ecm(-1) term. You can read my paper on this, see 'A convenient method ..... see my RG for correct title. The F-test suggested by PSS (2001) is affected by the number of the first differenced terms in the RhS of the model. Therefore, it is always much better to reinforce your cointegration using the t-ratio. Therefore, you have cointegration!
Arize, A.C. (2017)
"A Convenient Method for the Estimation of ARDL parameters and Test Statistics: reaiUSA trade balance real effective exchange rate relation" in IREF, 50, 75-84
Ok Chuck, but every paper I've read use the F-test but I'll read up your paper....thanks a lot!
....a digression: are you a Stata-user? If you are, I need some assistance with the Stata command for performing the "unit circle" test?
Ngozi, I am not saying you shouldn't report the F-test but simply pointing out its weakness. You obtained cointegration at 10% and that is good but with the t-ratio you found found a stronger evidence. You should report both as consistent evidence of cointegration. Please read my paper! I do not use Stata but I use the original program of Pesaran, Shin and Smith (2001). If you want me to help you run your data, you can forward it and my office will send you a quick rejoinder. We are on summer break and school opens on the 6th. If
you decide to do so, please just send the relevant variables only. Indicate the regressand.
I am unable to access your paper. I typed the title on Google and nothing came up, I typed on Google Scholar the link came up, though without a pdf file. I clicked on the link and got this error message:
"A problem was encountered providing the content you requested
Please try again later or contact our Customer Service team providing the below reference number.
Reference Number: 18.7bbf5d51.1496147890.11a4b4ce"
...I have no idea what the problem is. Is it possible you send the pdf?
Thanks.
Ha! So, how do they want us to read it if it's unavailable?...ok, thanks!
It is available through the library! You know how you get your journals over there. Are the entire papers on Google or Google Scholar? I f you send your email I could tomorrow attempt sending it to u assuming the contractual law allows I am doing work with the department of banking & finance Nsukka & Lagos universities and they all have access.
I'll try accessing through the University Library, if still unsuccessful I'll let you know.
what happens in a case where you do not have long run relationship at all levels of significance?
There are 2 ways to make a decision about cointegration the first is a t-test on the lagged error correction term, In softwares such as E-views that does not allow for different lags and plus the fact that the first-differenced terms in ARDL can cause FPSS to be insignificant, your problem often arise. & the t-test can be used.(I will be glad to write some more on this if you need more help). The second is the F test on the lagged level terms in the equation. Lets call it the FPSS. Therefore, when FPSS is insignificant you can use the t-test.
Dear Chuks, is there any theoretical evidence to support your assertion about t-test interpretation.
I am not st home presentlay but simply reading the first two pages of PSS (2001) gives you the answer to your questio.
I believe my answer is correct. Since you are interested in this area you may also want to read Neil R. Ericsson and James G. MacKinnon (2002) Econometrics Journal vol 5 285-318 and Anindya Banerjee , Juan J. Dolado and Ricardo Mestre (1998) Journal of Time Series Analysis Vol.19 , 274-279. You will really enjoy reading them.
dear all,
I'm running ARDL test, the result according bound test that there's no conintegration. However, the ECT was negative and significant, plus through the wald test I found there are short run causalitities of all regressors on the dependent variable.
thus can I conclude that there's cointegration (i.e. long-run equilibrium)?
@Dalia,
Since there is no cointegration, there's no long-run relationship hence no need to perform an ECM. Simply interpret the short-run results and maybe add causality tests to enrich your findings.
Hi, may variables are mix-stationary, I(0) & I(1). I ran ARDL bounds test and got results on long-run relationship. I want to now estimate the causality. How can I do that? Can I run VECM in this case since my data are mix-stationary? If not, what can I do to obtain the causality?
Dear Barua,
Given that you have a mixture of I(0) and I(1) variables that are cointegrated, you can estimate the causality using the VECM. Watching this You tube can be of help. https://www.youtube.com/watch?v=dnPd6yRAjXg
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation