Enders (3rd ed., p. 70): “Under the null hypothesis that all values of [the sample autocorrelations] = 0, Q is asymptotically chi-squared…If the
calculated value of Q exceeds the appropriate value in a chi-squared table, we can reject the null of no significant autocorrelations.”
So Enders actually states 2 versions of the null. At any rate, I’d argue that the word ‘significant’ should apply to the results of the test, not to the hypothesis
being tested. Otherwise, the distribution of the statistic under the null would depend on the significance level of the test, right?
I’d score this Allin 1, Enders 0. Now back to what I should be working on…
PS
From: gretl-users-bounces@lists.wfu.edu [mailto:gretl-users-bounces@lists.wfu.edu]
On Behalf Of Pindar
Sent: Friday, August 03, 2012 4:18 PM
To: Gretl list
Subject: Re: [Gretl-users] serial correlation test
Am 03.08.2012 00:13, schrieb Allin Cottrell:
On Thu, 2 Aug 2012, Alessia Via wrote:
So, in this case:
Equation 1:
Ljung-Box Q' = 0,67263 with p-value = P(Chi-quadro(4) > 0,67263) = 0,955
Equation 2:
Ljung-Box Q' = 1,25278 with p-value = P(Chi-quadro(4) > 1,25278) = 0,869
Equation 3:
Ljung-Box Q' = 0,0826623 with p-value = P(Chi-quadro(4) > 0,0826623) = 0,999
I have to reject the null of non significant correlation for eq. 2??
No. the p-value of 0.869 means that you are nowhere near rejecting
the null hypothesis (which, by the way, states that there is no
autocorrelation, not that there is no "significant"
autocorrelation).
Then we have to inform Prof. Enders that in at least the 2. edition of his book "Applied econometric time series" he is wrong by writing this.
See my correction to Pindar's posting at
http://lists.wfu.edu/pipermail/gretl-users/2012-August/007820.html
Allin Cottrell
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users