Coherence and phase

[Alberto]

Hi Tom,
I am trying to implement an exercise based on spectral analysis methods in the spirit of Sargent 1987 (Macroec. Theory, 2nd ed., sect 11. Alternative definitions of the business cycle) looking at spectrum, coherence and phase of various real GDP annual growth rate measures. I attach a sample programme looking at these measures for US, UK and Japan from 1961q1 to 2014q4 using the @spectrum and @crosspec procedures.
I have three main questions:
1) how can I save or display separately coherence and phase lead from the @crossspec procedure (there is no option to save them as for spectrum in the @spectrum, right?)?
2) how can I change the horizontal scale for the spectrum, say to have it in fractions of pi?
2) I am not sure what the best options such as ordinates, window, weights etc. are as they seem to differ across examples from Hamilton, Brock and Davis etc.: would you have any practical guidance?
thanks a lot
Alberto

[TomDoan]

Hi Tom,
I am trying to implement an exercise based on spectral analysis methods in the spirit of Sargent 1987 (Macroec. Theory, 2nd ed., sect 11. Alternative definitions of the business cycle) looking at spectrum, coherence and phase of various real GDP annual growth rate measures. I attach a sample programme looking at these measures for US, UK and Japan from 1961q1 to 2014q4 using the @spectrum and @crosspec procedures.
I have three main questions:
1) how can I save or display separately coherence and phase lead from the @crossspec procedure (there is no option to save them as for spectrum in the @spectrum, right?)?

I added options for that to the procedure. See

https://estima.com/forum/viewtopic.php?f=7&t=2404

2) how can I change the horizontal scale for the spectrum, say to have it in fractions of pi?

That's what it was already doing.

3) I am not sure what the best options such as ordinates, window, weights etc. are as they seem to differ across examples from Hamilton, Brock and Davis etc.: would you have any practical guidance?

A flat window with a width somewhere around the square root of the number of ordinates generally works OK unless the spectrum has very sharp features.

[Alberto]

Great, thanks a lot for your feedback.
Best regards
Alberto