Identification with 'too many' endogenous var's

In our experience it is possible to add more variables to the varobs list than there are exogenous shocks in the model when running the identification command. This is perhaps a simple error on our part which we ought not be making. But we are nonetheless surprised that this is possible. Should this not lead to an error message?

In any case we are also wondering what identification uses as endogenous variables if there are ‘too many’ variables in the varobs list.

Thank you for the help.

Sincerely,
Rob Luginbuhl

When there is stochastic singularity, it is still possible to estimate parameters and evaluate identification strength, albeit without using likelihood based methods. The identification toolbox, in that case, shifts to simulated moments and computes an ’ information matrix’ based on simulated moments uncertainty (see publications.jrc.ec.europa.eu/re … iv_312.pdf).

Stochastic singularity is not necessarly an error, since even such deficient number of moments is able to provide some information about model parameters.

This is the reason why no error is provided. Still you are right that perhaps a warning about that singularity can be added. In any cases, identification analysis can (and indeed has to) be done also in that case because, as I said before, stochastic singularity does not impede to perform some form of estimation/inference.