NON normality of residuals in panel data

[christina.sakali]

Hello all,

I am running a regression using panel data and 154 observations.

I am using the Panel Least Squares Method with fixed effects specification on Eviews.

The problem is I did a histogram normality plot and the Jarque-Bera statistic turned out to be significant (its value is around 60 with Prob = 0.000) which means that the residuals are NOT normally distributed.

What do you think I should do? Is there an estimation method that is more appropriate in this case? Or does this mean that the model and the estimation results are invalid?

I would greatly appreciate any responses as this is urgent.

Thanks a lot in advance.

[startz]

Ignore it.
If you have 154 observations normality is of very little importance.

[christina.sakali]

Thank you very much for this, I really appreciate your quick response.

Is there any reference (textbook or paper) where this issue is discussed, in case I will be asked to justify it.

[startz]

Basically, in the standard regression model the coefficients are asymptotically normal--subject to a few regularity conditions. Any graduate level econometrics text will have some discussion.

[christina.sakali]

Thnx a lot, I 'll look for this.

[christina.sakali]

Ok about the coefficients, but what about the DW statistic and the other diagnostic tests? Are these results reliable if residuals are not normally distributed?

For example I read in Baltagi (2001) 2nd ed. that Blanchard and Matyas (1996) studied the robustness of several tests for individual effects with respect to non-normality of the disturbances. Their findings suggested that the F-test is robust against non-normality while the LM & LR tests are sensitive to non-normality. Baltagi (2001) 2nd ed. Econometric analysis of panel data, p.72

This made me thinking about the diagnostic tests (especially the DW statistic) and whether we can trust their results in the case of non-normality of the residuals.

[haji1368]

Hi all
please help
please help in panel data in eviews 9 for normally data
n=191
T=3 Obs or nt=573 of company bours In test descriptive JB for all variable s is very much . with prob 0.0000
In test f prop 0.00000 and DW is .066 and r^2 is .33 and prob for
variable s is upper . 05!!
in test husman prob 0.0000 and prob for variable s is upper . 05!! R^2 is .79 and DW 1.55
with added, ar (1 ) for test... dw 1.78 !, but prob for variable s is upper . 05!!
Please help
Why? How normal data??
tanks