Simple RBC model setup error for bayesian estimation

Estimation ML/Bayesian estimation
144

Here is the dynare file and data file
RBC_v40.mod (4.5 KB)
data_Mar26.csv (537 Bytes)

Also, another question - can we change x-axis scales of prior-posterior density plots?

145

It seems you scaled your shock standard deviation in the model by 100 to have 1 represent 1 percent. However, in the data 0.01 is equal to one percent.

146

Thank you so much for your help prof. Pfeifer. Really appreciate it.

Also, my question : Can we change x-axis scales of prior-posterior density plots? Or can I find the data for plots in one of the estimated files and I can use that to create my own plots?

147

That is not easily possible, but the manual documents various options that may be helpful.

148

Thank you Prof. Pfeifer. I will check again.

Also, regarding moments - suppose for model moments- I get 2.2 as SD then that is equivalent to 0.22 which is close to the data moment which I calculate from data. However, after estimated model, I get 0.11 as SD - since it has been calculated using data, I will report it as it is? Correct or i am still missing something.

My point is simulated model and estimated model(bayesian) moments should match the data?

2.2 SD will be read as 0.02 SD for simulated model which is not close enough to data moment of 0.12?

149

From what I can see, your prior forces the standard deviation to be singificantly higher than in the data. Also, why does your observation for c_obs have a clear downward trend?

150

I chose informative priors close/ equal to the calibrated values
Yeah, I noticed the trend- I logged and detrended the original data (consumption divided by population). I have tried 2 methods to de-trend : removing the trend theough linear regression as well as one-way hp filter but I still get that trend.

151
  1. But the calibrated values may not make sense.
  2. Then something must go wrong in your detrending. A linear trend should clearly remove the downward movement.
152
  1. Calibrated values are chosen to what generally exist in literature
  2. I have checked twice but I will check again if that is where the problem exists.
  3. Also, I am not able to find forecasted values anywhere. I know they are not part of the folder oo_ structure but I also could not find it on the hard disk. Where exactly should I look.
153
  1. That may well be, but those values may not be suitable for your country/data.
  2. Can you provide the code you use for detrending?
  3. Did you set the forecast option of estimation?
154
  1. I will look into this.
  2. Sure, but I detrend in python- will it be okay for you to look at python codes?
  3. Yes, I used forecast option of estimation.

Thank you gain for al your help Prof.

155

The figure clearly shows that your detrending did not work correctly:

image
image574×549 15.7 KB

156

Can we extract the following from output files:

  1. Values which are used to plot Bayesian IRFs ?
  2. Values of metropolis output for all parameters?
  3. Values which are used to plot the parameter posterior distributions (i.e posterior plot values)?
  4. Is there a better way to study moments - bayesian v/s frequentist v/s data than comparing point estimates?
157

1.-3. The posterior draws used should be saved in the metropolis subfolder in the FILENAME_posterior_draws*.mat files.
4. Can you elaborate?

158

Thank you Prof. Pfeifer for 1 to 3.
On 4, I want to ask whether moments can be generated from the entire posterior distribution of paramters and not just mean of the posterior estimates of parameters?

159

That is what the moments_varendo option is for. See the manual.

160

I read through the manual and various posts. I have two clarifications:

  1. Using “moments_varendo” option, I get Covariance and correlation. Where can I find the interpretation as to what they represent in greater detail- I read through the posts and manual. I still do not understand completely.
  2. Suppose, I am looking for the posterior mean of the of the theoretical moments of the endogenous variables, then how do I get those - is it stored under " PosteriorTheoreticalMoments.dsge.covariance.Mean.endogenous variable". I am still unclear about the results of varendo and what exactly each result denotes? Can I plot the posterior distribution?Looking forward to your help.
161

In other words, is it possible to get posterior mean of the theoretical means for each endogenous variable? I understand that **PosteriorTheoreticalMoments.dsge.covariance.endogenous varaible**Contains the posterior mean of the variance-covariance matrix and Diagonal elements, like .consumption, represent the posterior mean of the variance of consumption. However, if I want the mean of an endogenous variable from the entire posterior distribution of parameters and not only mean, how do I get that. As of now “moments_varnedo” only reports “correlation” and “covariance”

162

If you are interested the posterior distribution of the variables means, they are reported when you run the smoother option.

163

Thank you for your response and help. So I ran the smoother option - are the results stored in “SmoothedVariables.Mean”. I can use those values to plot the posterior distribution of means? Will it be correct to report or analyze - mean of all the means which are reported under “SmoothedVariables.Mean”?