Lag/lead correlations in Cross Correlogram

[viviancwj]

I am using Cross Correlogram with a lag of 10 to identify the highest correlation between two variables. My understanding is that the figures in the tables are the correlation between the two variables (with lag/lead as indicated by i). However, I noticed that other than i=0, the correlations of lag/lead variables do not match exactly the actual correlation -- if I generate the lag variable and calculate the correlation. Could anybody explain to me why? Thank you.

For example

Here is the result of Cross Correlogram:

Date: 03/26/15 Time: 17:12
Sample: 1995Q3 2014Q4
Included observations: 74
Correlations are asymptotically consistent approximations

Q3_9,CILOANYY(-i) Q3_9,CILOANYY(+i) i lag lead

. |**** | . |**** | 0 0.3789 0.3789
. |*** | . |***** | 1 0.2770 0.4601
. |** | . |***** | 2 0.1588 0.5059
. | . | . |***** | 3 0.0432 0.5008
.*| . | . |***** | 4 -0.0530 0.4798
.*| . | . |**** | 5 -0.1241 0.4264
**| . | . |**** | 6 -0.1773 0.3585
**| . | . |*** | 7 -0.2057 0.2849
**| . | . |** | 8 -0.2227 0.1978
**| . | . |*. | 9 -0.2238 0.0905
**| . | . | . | 10 -0.2187 -0.0220

If I genr q3_9_lag1=@lag(q3_9,1), and correlate q3_9_lag1 with ciloanyy, I got 0.454389, which is different from 0.4601 in the above table.

[EViews Glenn]

We discuss this in our documentation of the ordinary correlogram. Following the literature, we employ the computational shortcut which uses a common mean across all calculations:

Note that the autocorrelations estimated by EViews, which follow those in the literature, differ slightly from the theoretical definition of the autocorrelation...
EViews employs the same overall sample mean as the mean of both [Y_t] and [Y_{t-k}]. While both formulations are consistent estimators, the EViews formulation biases the result toward zero in finite samples.

Note that the output indicates that we are using the asymptotically consistent approximations.

[viviancwj]

Thank you, Glenn. Could you please point me to the documentation of the ordinary correlogram you mentioned in your post? Thanks.

[EViews Glenn]

Go to User's Guide 1 and look in the index for correlogram:autocorrelation function.

It's on page 393 of EV9 Manual. Don't know what page in earlier versions but should be around there.

[tpk1001]

Hello,
I am wondering if there is any way to test the significance of the correlation coefficients provided in the eviews cross-correlogram (I understand the asterisks are a proxy for this).
It seems this is not possible in eviews 7 which I am using. If that's the case can I use the following test instead? http://vassarstats.net/textbook/ch4apx.html.
A second - more general question. Is there any rule on what is a strong or weak correlation? The above test will only tell me whether the coefficient is relevant on a certain significance level
but I also need to assess the level of correlation....

Thanks for your help.