OSR for average loss across parameter space/posterior


I’m interested in exploring OSR under parameter uncertainty. I would like to find the OSR coefficients where the loss function is the average loss across some distribution of other parameter values. For example, it could be the average across the posterior distribution for the non-policy parameters. Or I could just vary one other parameter of interest and consider the average loss across its values.

In order to do this, my current thinking is that I could modify the in-built Dynare function osr_obj(), so that it computes the average loss across some parameter distribution for a given set of OSR coefficient values (instead of the loss for a given parameter setting). I could then replicate Dynare’s OSR function, using this modified osr_obj() instead of the original.

I would be interested to know if anyone is aware of any pre-existing papers or code that conducts a similar exercise using Dynare, or whether my suggested approach is a sensible/feasible way of doing it.


Yes, that sounds feasible. Essentially, you are integrating over the parameter distribution.

1 Like