ARDL, how to solve serial correlation problem.

[Elshan Alakbarli]

Dear all,

I am new to using Stata, and I am trying to find out the effect of energy consumption on economic growth. As a dependent variable I am using GDP, and as indep. variable Energy consumption, GFCF and Labour force data from 2000-2019. While running ADF test, I found out I(0) for GDP and I(1) for the indep. variables. As a next step I am trying to use ARDL model, however, I am not sure how should I proceed. When running bound test I am getting F-statistics value which is higher than upper bounds, but on the other hand t-statistics value which is lower than lower bound levels in all significance level. So, this confuses me a lot. I am not sure what should I do. While running diagnostic test, (Breusch Godfrey), I get a serial correlation result..
Could you please, help me on solving these issues? and also could you give me an advice on what steps should I follow if I am missing any?
Thanks a lot for your time and help in advance!

[Sebastian Kripfganz]

We are discussing suggested steps and the interpretation of the bounds test results in our documentation of the ardl command:

• Kripfganz, S., and D. C. Schneider (2022). ardl: Estimating autoregressive distributed lag and equilibrium correction models, Manuscript.

If the t-statistic does not reject the null hypothesis, then this is evidence against a long-run relationship, even if the F-statistic rejects the null hypothesis of no long-run relationship. Serial correlation is usually dealt with by increasing the lag order of the variables. If you used the BIC as a model selection criterion, try the AIC instead.

What is the frequency of your data? If you have annual data, then I am afraid 20 observations are unlikely to be enough for obtaining reliable results.