Estimation, IRFs and variance decomposition

[jteres]

Hello,

I'm beginner to Dynare and I would like to ask a few (probably basic) questions to make sure whether I am not wrong:

1. Subsample estimates: I estimate the model parameters over the full sample and in two subsamples. Is it possible to obtain the full sample estimates that lie outside the interval created by the subsample estimates? E.g. for 2000-2007 I obtain the estimate of 0.56, for 2008-2013 I get 0.61 and for the whole 2000-2013 period 0.65. I expected that the estimate over the full sample should be somewhat average of the subsample estimates...

2. IRFs after the estimation: In case of estimated model can we use stoch_simul or should we only use bayesian_irf command to obtain impulse response functions?

3. Conditional variance decomposition: I would like to generate the forecast error variance decomposition at different time horizons. I use the command conditional variance decomposition. Just to make sure: is it correct and is that how we obtain the forecast variance decomposition?

4. The output of conditional variance decomposition: When I use conditional variance decomposition as stoch_simul option Dynare prints a nice table with the variance decomposition. In case of the estimation command conditional variance decomposition option no table shows in command window and the output is stored in oo_ file, but its form is not very nice. Can we translate it somehow to obtain such a plausible printed result like in the case of stoch_simul variance decomposition?

5. Historical decomposition: When I run shock_decomposition, I obtain the graph. Is it possible to change the colours (e.g. from rainbow ones into black and white shades)? Can we add years/quarters on the X-axis?

6. Reproducing the output of historical decomposition in Excel: If what I asked above is not possible, can we reproduce the graph that Dynare shows in Excel? I know this question appeared earlier, but I have not found satisfactory answer. I know where to find the output in oo_ file, but its form is difficult to work with (three dimensional).

Could you please help me?

Thank you very much in advance for the answers!

Best regards,
Jan

[jpfeifer]

1. Subsample estimates: I estimate the model parameters over the full sample and in two subsamples. Is it possible to obtain the full sample estimates that lie outside the interval created by the subsample estimates? E.g. for 2000-2007 I obtain the estimate of 0.56, for 2008-2013 I get 0.61 and for the whole 2000-2013 period 0.65. I expected that the estimate over the full sample should be somewhat average of the subsample estimates...

This is possible but should be unlikely. But it looks as if you are only looking at point estimates. You should not do that. An indication of problems would be when the credible sets/HPDIs of the estimates are disjoint.

2. IRFs after the estimation: In case of estimated model can we use stoch_simul or should we only use bayesian_irf command to obtain impulse response functions?

See http://www.dynare.org/phpBB3/viewtopic.php?f=1&t=4650

3. Conditional variance decomposition: I would like to generate the forecast error variance decomposition at different time horizons. I use the command conditional variance decomposition. Just to make sure: is it correct and is that how we obtain the forecast variance decomposition?

Yes, that is correct. However, keep second point in mind, i.e. the difference between the FEVC at the mean parameter (stoch_simul) vs. mean FEVC (estimation)

4. The output of conditional variance decomposition: When I use conditional variance decomposition as stoch_simul option Dynare prints a nice table with the variance decomposition. In case of the estimation command conditional variance decomposition option no table shows in command window and the output is stored in oo_ file, but its form is not very nice. Can we translate it somehow to obtain such a plausible printed result like in the case of stoch_simul variance decomposition?

No, you would have to do that yourself.

5. Historical decomposition: When I run shock_decomposition, I obtain the graph. Is it possible to change the colours (e.g. from rainbow ones into black and white shades)? Can we add years/quarters on the X-axis?

6. Reproducing the output of historical decomposition in Excel: If what I asked above is not possible, can we reproduce the graph that Dynare shows in Excel? I know this question appeared earlier, but I have not found satisfactory answer. I know where to find the output in oo_ file, but its form is difficult to work with (three dimensional).

You would also have to do that yourself. Regarding the three-dimensional array, the manual for 4.4 documents the entries. Alternatively, you could manipulate the graph_decomp.m of Dynare. The critical parts are the fill-commands like

Code: Select all

                fill([x(i) x(i) x(i+1) x(i+1)],[yp yp+zz yp+zz yp],k);
                yp = yp+zz;


where the second bracket provides the colors for filling.

[jteres]

Thank you very much for the answer and helpful comments!

Kind regards,
Jan