test for autocorrelation when estimating ARCH-Model

[Michal83]

Hi, I estimated in EViews 8 an equation using the ARCH-Method. In residual diagnostics, there is no Breusch-Godfrey Serial Correlation LM. Why?
How can I test for the autocorrelaction? In my equation, I use a lagged endogenous variable, so the Durbin Watson test is biased (and the Breusch-Godfrey test is not offered). Is there any easy way, how to calculate the Durbin h-statistic? Or other tests? Thank you very much in advance.

[strypste]

open your equation object and go to:

view -> Residual diagnostics -> Correlogram - Q-statistic

and then click "ok".

The output will show the Ljung-Box Q-statistics and its p value. The null hypothesis is that there is no autocorrelation (See http://en.wikipedia.org/wiki/Ljung-Box_test)

[forever]

Hi，strypste，could one can test the autocorrelation by clicking Proc/make residual series and then clicking view/correlogram？which one is standard？

[strypste]

Hi

I think this depends on the model you estimated.

For example:

Model 1: y c ar(1) ar(2)

Model 2: y c y(-1) y(-2)

You will see that for model 2 the two methods will show that same results, but for model 1 they will be different. In particular, the method you propose will not have p-values for the first two lags. This has to do with the degrees of freedom. Somebody else and I actually already posted a question asking why exactly this is the case. (http://forums.eviews.com/viewtopic.php? ... 718#p12565)

[trubador]

In particular, the method you propose will not have p-values for the first two lags. This has to do with the degrees of freedom.

Yes, that is correct. When the residuals from an estimated ARIMA model are used, degrees-of-freedom are reduced by the number of estimated AR/MA coefficients.

[forever]

Dear strypste：
I find the two test ways are conflicted in my equation,now I think the first way is correct when you fit a ARIMA model,such as what have done in "Forecasting the Volatility of Philippine Inflation using GARCH Models". Do you agree with me? As to your question(viewtopic.php?f=18&t=3718#p12565),it should be occured in the first test way.I don't know the true difference between model 1 and model 2 all the time，as I know, the model 1 should be AR(2) model,but I don't know what mean about the model2?.Could you tell me?

[strypste]

This post explains the difference between model 1 and model 2:
http://forums.eviews.com/viewtopic.php?f=7&t=465

[forever]

Thank you vary much，strypste :)