Dear Researchers
I have estimated the Supply Function for ready-made garments exports but the Jarque-Bera value of estimated model is so large. Does the normality assumption is basic requirement of ARDL model or In ARDL methodology this assumption is not required.
thanks in advance
According to Dave Giles, normality assumption is not so important. The main worry shoud be the autocorrelation. (http://davegiles.blogspot.ro/2013/06/ardl-models-part-ii-bounds-tests.html).
Regards!
Maybe you should try to consider a heteroskedastic ARDL model first. Normality is a maintained hypothesis, so if that hypothesis does not hold, it clearly points out a specification problema.
Dear Carlos
I did not get your point. can you explain it again?
OK. Every maintained hypothesis in a model should be tested in order to justify it. If all the assumptions behind such a model (error normality, error non-heteroskedasticity, error non-auto-correlation) are validated by the data used for the estimated model, it will be said that this model has no specification problems or that this model is statistically valid. There are many other tests one should also try, for instance, the absence of endogeneity problems, but the more compelling tests are the normality/non-heteroskedasticity/non-auto-correlation tests. Otherwise the model is not so good, although usable. This is really the standard time-series econometrics principle since David Hendry's "bible", DYNAMIC ECONOMETRICS. This is the most important book any serious econometric practitioner should read (and then apply)! After reading it, the issue of model mispecification will be clear BUT then the problem is to use a really good set of tests (meaning powerful tests) which of course are not quite easy to implement, specially for multivariate time series models (up to my knowledge).
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics