Some question about estimation and out-of-sample forecast

Estimation ML/Bayesian estimation
1

Hello everyone,
I am a beginner in dynare. I encountered some issues while exploring the functionalities of dynare. I attempted to find answers through the user manual but was unsuccessful. The original model is from Prof. Pfeifer’s Github, and I am using dynare version 5.3.

original code and data:
ls2003.mod (1.7 KB)
data_ca1.m (6.5 KB)

(1) I only retained the prior settings for the standard deviation of shocks under estimated_params. I noticed that modifying the prior distribution, mh_nblocks, mh_replic, and mh_jscale for the standard deviation of shocks affects the estimation results, but does not impact out-of-sample forecasting results. Is the situaiton reasonable? Why doesn’t it affect out-of-sample forecasting results? Does this imply that setting different shocks for the same model doesn’t affect out-of-sample forecasting results, meaning out-of-sample forecasting results only depend on the setting of endogenous variables in the model?
ls2003_q1.mod (2.0 KB)

(2) By removing the estimated_params command and using calib_smoother, I am still able to obtain out-of-sample forecasting results, and they seem to be exactly the same as the results in question 1. Why is it possible to run the model using calib_smoother instead of the estimated_params command? Am I essentially instructing the estimated_params command to perform only out-of-sample forecasting?
ls2003_q2.mod (1.9 KB)

(3) Can the variable estimation results obtained after using the calib_smoother command be used to assess the model’s fit to the data, such as comparing the standard deviation of the output from calib_smoother command to the standard deviation of the data?

Since English is not my native language, I ask for your understanding if my choice of words or expressions may cause any discomfort. Sincerely, I appreciate every researcher willing to help me with my questions!
Hongming Zhang

2

There are a couple of interrelated issues. At first order, there is certainty equivalence. The shock variances will not affect the decision rules per se. However, the standard deviations will affect the likelihood and therefore the estimate of the other deep parameters. That in turn will affect the results from the smoother. But in your application, you don’t estimate the fundamental parameters. Thus, the decision rules underlying the smoother estimates will always stay the same. That is also the reason why it does not matter whether you run the calib_smoother or an estimation-command. Both result in the same decision rules and same smoother results.

3

Thank you for your clear and understandable explanation! After reading your response, I believe I now understand the principles involved in these issues. Thank you once again sincerely! :smiling_face_with_three_hearts: