Hi, I am wondering if there is any theorem or principle to support the estimation of DSGE. It’s great to run a code and get some feedback, but still need to calibrate the parameters if I want to get something different. And again if I want to add new structure on old one. More likely to be blackbox for me instead of rigorous programming(although it is).
My question is if any theorem to help us brief or control the final result like math subject in a very basic structure? For example, it’s fantastic to have a principle in RBC model can tell us the final results given the parameters. Then we can develop to more sophisticated one without loss of readability.
Thank you for your reply!
Sorry, but I don’t understand what your question is.
Dear jpfeifer,
Sorry for delayed response. I mean if there is any theoretical approach for us to get the control before we run the code. For example, we calibrate depreciation rate, share of capital sort of things, then we run the code based on calibration and model’s structure. Can we have some estimation like upper bound estimation for parameters, claiming “based on the model and parameters, consumption adjustment peak for TFP shock won’t more than 3%” before we run the code. Hence we can use code to verify our claiming.
Thank you so much!
No. The closest you can get is by performing a prior predictive analysis, i.e. compute the statistic of interest for draws from your prior distribution (which reflects your idea about likely parameter values). That being said, the posterior by construction cannot have a support different than the prior.
I see, and reason we use Bayesian to estimate data is that property of high dimension and nonlinear models of DSGE. So do MLE as they all adjust parameters to fit data best. And calibration just test specific parameters’ performance of model. In all, there is almost impossible to get theoretical bounds without help of numerical algorithm. Am I right?
You are asking for
Those are objects you define before seeing the data. Either you have an idea about plausible parameter values or not. The data can only show you whether these values are considered likely.
Yes! Thank you for your help to figure out my problem!