I have done identification strength/sensitivity check for all the estimated parameters in my DSGE model, please refer to the attached pdf graph and please have a look. I find identification strength in upper graph and sensitivity component in lower graph for the first parameter are very different, I have shown iet to another DSGE person and that person also find the difference a surprise, because both identification strength and sensitivity depend on log likelihood with respect to the first parameter. We know formulas for identification strength and sensitivity component are different, however since they all depend on curvature of likelihood with respect to the parameter, why are they so different? (we mean why the first bar in upper graph and the first bar in lower graph look so different?
Thank you very much.
JesseIdentification strength and sensitivity.pdf (165.4 KB)
Sensitivity is only one of two component of identification strength. If there is high collinearity with respect to the effect of other parameters on the likelihood, you will get a picture like yours. Taken alone, the parameter strongly affects the curvature of the likelihood. But considered jointly with the other parameters, it is hard to isolate the effect of that one parameter as the other parameters have a similar effect
Thank you very much, Johannes, since there is high collinearity with respect to effect of other parameters of the likelihood, does it indicate weak identification?
Thank you again and look forward to hearing from you.
The identification strength measures the overall effect. There you can see that the first parameter is not that well-identified.