I have Total factor productivity as dependent variable and other 4 varibles as regressors (R&D domestic stock, R&D spillovers and GAP) i introduce first difference to demonstrate the annual growth (mentionned in the theory). i find that all variables are stationary in this difference. but i do no how to consider them , i mean if my variables are integrated in level or I(1) because my model is already with first difference (endogenous growth theory).
My question is how can i check if their are autocorrelation and heterscedasticity in my panel using Eviews?
and what is the pruposal solutions (knowing that all variables are not cointegrated (using panel cointegration test))
My panel is already with log since TFP is residual of cobb douglas. Crossid number 28 country and time T=12 ans (with difference)
Thank you
Hi not a real expert but here goes:
There are specific autocorelation tests but you can check for serial correlation the easy way just regress the variable in question to its own lag-ed (t- or t-2) value, if the p value from the regression is significant then you have autocorelation, but since you have panel data I think this would not do it
You can check for heteroskedastisity also the easy way by making a plot of the residual vs fitted values. If the residuals exhibit a pattern like a cone or whatever it is heteroskedastic. If the residuals are randomly dispersed all over the place, above and below, the fitted line you are ok.
I am a beginner so I welcome more insight and answers as to enlarge my knowledge also.
I like the model BTW.
Kind regards
Dear Samia,
In case of Panel data, you can use STATA. In STATA, you can check the auto correlation through wooldridge test (xtserial), and heteroskedasticity through Breusch-Pagaen test (hettest).
In case of EVIEW, when you run the models in the software, the results itself shows the auto-correlation and heteroskedasticity values in the model result itself. Please check.
Dear Brahmadev can you tell me more about how the results of estimation in eviews show the autocorrelation and hetersc please
Dear Samia, when you will run the panel OLS in EVIEWS, the final output sheet will show the Durbin-Watson test results. Durbin-Watson test indicates the auto-correlation issue, if the test result is within the limit of 1.8-2.2 then there is no auto-correlation problem.
Next you can run the white test for heteroskedasticity, if the the p value of the test is significant then there is a heteroskedasticity problem, if p-value is insignificant then there is no hetero problem.
thanks
Thank you dear Brahmadev but in eviews their is no White test for heteros for panel. Thank you all
Dear Samia, once you run the pooled OLS for the panel data, you can then proceed to the view tab. Under the view tab you will find Residual Diagnostics, where you can run heteroskedasticity test. Hope this is helpful.
Thanks
You can check the existence or otherwise of autocorr and hetero by going through the residual diagnostics after the regression.
All the best.
Dear Amit Tripthy and Samuel Tawiah thank you very much, i effect the test and i find heteros, just want to ask about FGLS does it solve the problem of heteroskedasticity?
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series