Dynare forums

Hello everyone,

I am replicating one paper and I attached my mod file. I have some questions regarding the bayesian estimation and forecasting:

First Question: In the mod file authors fist set the initial values for the parameters and then start for a bayesian estimation with setting different initial value. I don't understand this. For example; they initially set "rho_a = 0.85" and then for the bayesian estimation "rho_a, 0.94 , 1E-8, 0.999, beta_pdf, 0.80, 0.10;". What is the logic of setting the begining value for parameter and making the bayesian estimation at the same time?

Second Question: I want to make forecasting but I have data problems. Can I simply take the results of the bayesian estimation like taking the posterior means and set them as initial value and make forecasting? Is it the same as making the bayesian estimation and making forecasting afterwards?

Third Question: I want to compare my forecasting results with simple VAR model forecasting for example. But when I use "forecast" command it just gives me some graphs, which command I should use to receive proper result so that I can compare it with my VAR forecasting?

Thank you so much!

1. You should always try your model with calibrated parameter values before doing estimation to see whether the model works. The starting value should in principle not matter if you do estimation correctly.
2. What do you mean with "data problems"? And: No, you cannot do this. Forecasts at the end of the sample need to start at the smoothed values for the state variables, not at the posterior means. Regardless of whether you do Bayesian or classical forecasting, you need to solve the data problem. The main difference between the Bayesian forecasts and the forecasting after estimation is one of interpretation and concepts (mean forecasts vs. forecasts at the mean)
3. The answer depends on whether you want to compare it to Bayesian or classical VAR. In the former case, you need Bayesian forecasts, in the latter one, the forecast command with "classical" forecasts will suffice.