Dynamic Panel Data Residual tests

[ZCF]

Dear All,

I'm done with my dynamic panel data inclusive of the sargan test. But I'm wondering if someone could help with Arellano & Bond first and second order serial correlion residuals test. How they're done in Eviews?

Thanks in advance,

Fernando

[EViews Gareth]

There is no automatic way to calculate these in EViews, but someone might be able to suggest a way to calculate them manually.

[kizil]

Hi all,

The user's guide actually has an example on testing for residual autocorrelation (p. 555). I assume that the procedure described there is also applicable to dynamic panel estimations. But there is one point that could not understand in that example: It reads: "Under the null hypothesis that the original idiosyncratic errors are uncorrelated, the residuals from this equation should have an autocorrelation coefficient of -0.5."

Could anybody tell me where does this -0.5 come from? And another question: In the estimation output there is this prob coefficient of 0.0814. Could I take this as the probability of the null hypothesis (there is first order serial correlation) being accepted? Is this how the test for residual correlation conducted?

Your urgent help is highly appreciated. Thank you.

[EViews Glenn]

Under the assumption that the data are originally uncorrelated, the MA(1) that you get when you compute first differences gives you a correlation coefficient of -.5.

E((e_t - e_{t-1}) * (e_{t-1} - e_{t-2}) = -sigma^2
E(((e_t - e_{t-1})) = 2*sigma^2

You should ignore the .0814 Prob in the estimation output and instead look at the Wald test results since those are for the null hypothesis of interest.

The Wooldridge book from which we take this example will do a better job of explaining all of this.

[durbin-watson]

Hi,

since EViews does not offer Breusch-Godfrey tests for autocorrelation for panels, I need to test if autocorrelation remains after having included a lag of the dependent variable as a regressor.

With the Durbin h-test, the formula says you have to get the variance of the lagged dependent variable (along with the normal DW stat given in the output and the number of observations). What if I have included 2 lags, which variance do I take to calculate the h-statistic, the one of the first lag or of the second?

Thanks.
Durbin

[EViews Glenn]

In the dynamic panel data context, I wouldn't even mess with any of the traditional measures. Arellano and Bond (1991) directly address this issue...

[pimenteljm]

Hi!

I am using a simple panel data OLS with 5 cross-sections and 41 periods.

I am using a lagged dependent variable (all the other variables enter as levels), but I am afraid I am getting serial correlation. However, eviews doesn't provide an alternative to the Durbin-Watson. I tried to regress the residuals on the lagged residuals, but I am not sure whether it is the right approach.

Can you help me?

thanks in advance!

[d952]

Hello every one

I am in the same problem as well! I used Arenallo-Bond panel GMM estimator however, I can not find any way to do first and second order auto correlation test with Eviews, I would appreciate alot if some one tell me is it possible to do it manually with Eviews?

Thanks every one

[EViews Gareth]

Use the EViews 8 Beta.
http://www.eviews.com/EViews8/EViews8Beta/index.html