Calculation of theoretical moments with measurement errors

I would like to know if (estimated) measurement errors are included in the calculation of theoretical moments, e.g. variances, of a model when I run stoch_simul.


No, it isn’t. But we could add it if you juge it useful. We need also to make sure that it is possible to declare a measurement error in the shocks statement using simple a stderr on an endogenous variable.




Just out of curiosity, how do you include measurement errors in your model?

What I have done is as follows. Say output, y, is an observable for my model, then somewhere in my model declaration step I add the line

yobs = y + nu_y;

where yobs is the observed y and nu_y is my intended measurement error.

Is that how you do it? If not, how do you incorporate measurement errors?


Hi Karl,

If you follow this approach, obviously the reported variance for yobs will include the theoretical variance (ie the variance of y) and the variance of the measurement error (ie the variance of nu_y).

But, when you estimate a model with measurement error you do not need to add an equation in the model (which may be a bad idea because this will increase the size of the state vector). You can simply declare a prior on the variance of any observed variable (see the manual), then Dynare will understand that you want to add a measurement error on this variable. In this case, the reported variance for the variable is the theoretical variance (without measurement error).