Negative score in identification check

Identification/Sensitivity
1

Dear Jonas,

May I have a query please regarding the identification check picture (attached): am i right that, if a parameter has a negative score, it still means it is identified (though compared to those more on the right it is not as strongly identified)?

identification%20check
identification check.jpg1440×787 92.4 KB

I have checked through the forum and also your document regarding dynare graphs. The impression I have (and as it is documented) is that, as long as the scores in the upper and lower graphs are not BOTH zero, then that parameter is identified. Your document did mentioned a case when the (log) score is zero, but i am not sure what it means when the score is negative? Especially, I thought that as the score gets bigger and bigger (eg from very negative to less negative, and then to possible), it becomes more and more identified. But why zero is a special case of non-identification?

Thank you very much for your help!

Regards,
Zhirong

2

Yes, as long as a bar is displayed, the parameter is identified. Only when no bar is shown the parameter is not identified at all. Identification strength ranges from 0 to plus infinity. By using a log scale, the domain now is (-\infty, \infty). Values between 0 and 1 in terms of identification strength will therefore show up with log values that are negative. A 0 values in the level will be minus infinity in the graph, which is the reason no bar is displayed as you cannot plot infinity.

3

Dear Jonas,

Thanks very much for your quick reply. This is very clear and helpful!

Regards,
Zhirong