I’ve been working on a method for estimating a DSGE model and doing inference on parameters, and it seems to work pretty well, for a fairly simple model (https://github.com/mcreel/SimulatedNeuralMoments.jl/tree/main/examples/DSGE). It’s fast enough so that Monte Carlo is possible. I’d be interested to see how other estimation methods compare, either for the model I used, or for some other model, and I’m hoping people might be willing to propose models and perhaps share results obtained using other methods. This sort of game might help to develop use of Julia for estimation/inference methods for DSGE models, and we might also learn more about how reliable are the methods that are used. Any interest and/or ideas about how to run such a game?
1 Like
What is DSGE? Not everyone knows, neither do I.
1 Like
Sounds interesting. I’d guess you’d want to fix the model and the inputs and only focus on the estimation methods for this purpose.
My way of thinking of a standard problem would be: (1) a model that can be used to generate data, given parameters, (2) a prior or parameter space for the parameters, and (3) a design vector of true parameters. Given those things, people could apply different methods to estimate the parameters and compute confidence (credible) intervals. Then, by Monte Carlo, RMSE, confidence interval coverage, and other measures of interest could be computed, using whatever method the user likes. If the standard problems were agreed on, then different users could report their results. Perhaps all of this could be put into a git repository, to facilitate and centralize it, if there were interested users.