Serial correlation tests - Wald vs LM in Eviews
Posted: Wed Mar 28, 2012 3:54 am
Hi,
I have what might seem a rather basic question:
I'm testing for serial correlation for several regressions that I've run. I get conflicting results for the F-statistic and chi-square statistic - the latter always more likely to show statistical significance than the former. I understand that sample size is an issue and I note that in some instances, the F-stat (and t-stats for individual lagged residuals) remain statistically insignificant whilst the chi-square stat does not as I test higher order lags.
I'm wondering what others would do - stick to the F test (if sample size is small -but what would be "small") or take it that serial correlation is present whenever the chi-square test is significant. I'm also concerned that Eviews refers to the F-test as an "informal" test.
Many thanks.
M
I have what might seem a rather basic question:
I'm testing for serial correlation for several regressions that I've run. I get conflicting results for the F-statistic and chi-square statistic - the latter always more likely to show statistical significance than the former. I understand that sample size is an issue and I note that in some instances, the F-stat (and t-stats for individual lagged residuals) remain statistically insignificant whilst the chi-square stat does not as I test higher order lags.
I'm wondering what others would do - stick to the F test (if sample size is small -but what would be "small") or take it that serial correlation is present whenever the chi-square test is significant. I'm also concerned that Eviews refers to the F-test as an "informal" test.
Many thanks.
M