Question about Serial Correlation LM Test

[DDECalls]

I have a non-stationary variable, let's call it closing for this example. I am differencing closing to make it stationary so that it can be used in the ARIMA model after the unit root test.

After differencing, does this variable need to reject the null hypothesis Does not contain serial correlation in the Serial Correlation LM Test to be put into the model?
(Example code E-Views 10: d(close) c d(close(-1)), I put this equation in the Serial Correlation LM Test)

Or can I continue to apply the model if the ACF/PACF lines in the correlograms exceed the boundary. How does this affect model integrity or health?

 

[markbogorad]

EViews is great, always wondered why it's not used as much in the US as in the UK.

does this variable need to reject the null hypothesis Does not contain serial correlation

Need to reject the presence of serial correlation otherwise your estimates are going to be inefficient (not BLUE under Gauss-Markov assumptions). So if the null of your test is "does not contain serial correlation", you want to fail to reject this and a high p-value is what you're looking for.

Example code E-Views 10: d(close) c d(close(-1)

If you still have serial correlation, it might be that you're regressing your dependent variable on its own lagged value only and have no other explanatory variables present. These two likely have correlation, hence why your model might not pass serial correlation tests.

Maybe try d(close) c ar(1)? This would be using the ARIMA terms that you seek and might help with the serial correlation. Also uses the maximum likelihood estimator instead of OLS which is better.

I'm not sure the exact details of your project, but if you're just looking to see if lagged values are significant I'd recommend you try the ARDL model in EViews. It will show you the highest order of significant variables for each explanatory variable and the dependent as well.