I have studied your"An introduction to Graphs in Dynare" for several times,and I hope I could seek some help from you about explanation of shock_decomposition figure.

Taking your code of “Real Business cycles in emerging countries” as an example. If I want to see what shocks mainly drives Argentina’s historical fluctuation of Output, I would use shock_decomposition command after estimation command as following:

estimestimation(datafile=data_argentina, xls_range=G2:J107, logdata, mode_compute=6, moments_varendo, mh_nblocks=1,mh_replic=2000000, mode_check);
shock_decomposition y g_y;

Question 1:
Since y is denoted "detrended output " and the model is loglinearized by “loglinear” command before estimation, then the **vertical axis **in the figure of shock_decomposition to y means: (logy-logybar) =(y-ybar)/ybar,both of which are “Percentage deviation of detrended output from its steady state ybar”. Is that right?

Question 2:

Since g_y is denoted “Gross output growth rate” and the model is loglinearized by “loglinear” command before estimation, the the **vertical axis **in the figure of shock_decomposition to g_y means : logg_y-log((g_y)bar) which is percentage deviation of gross growth rate from gross growth rate trend, also means the deviation of net growth rate from net growth rate trend. Here, (g_y)bar is gross growth rate trend=gbar=1.0107 in data. Is that right?

Question 3:

It seems that some people report shock_decompositon y figure, while some report shock_decomposition g_y figure. I am wondering which one is better ? or Does it depend?

I think it’s yes.
With loglinear options, you log-linearize the model around the steady state. So the otputs are percentage deviation from the steady state.

I’m not very familiar with this literature, so I’m not 100% sure that what you say is correct. In particular I’m not sure of the analogy between gross and net rates.

3)I think it depends on what you whant to claim/explain, but if you what to deal with output I think the graph of y is better.

Yes, due to the loglinear option, you will have percentage deviations from the steady state.

Yes, g_y is the gross growth rate in that file. Due to loglinear, it will become the net growth grate. Again in deviation from its mean (which is approximately 0).

As Massimo said, it depends on the economic content you are after. Often people look at growth rates because this growth rate has an immediate correspondence to an object in the data. This is less so with the detrended output from the model. We do not really observe this/it is not easily computed as we cannot spearate the trend in the data given the assumptions of the model on what constitutes the trend.

[quote=“jpfeifer”]1) Yes, due to the loglinear option, you will have percentage deviations from the steady state.
2) Yes, g_y is the gross growth rate in that file. Due to loglinear, it will become the net growth grate. Again in deviation from its mean (which is approximately 0).
3) As Massimo said, it depends on the economic content you are after. Often people look at growth rates because this growth rate has an immediate correspondence to an object in the data. This is less so with the detrended output from the model. We do not really observe this/it is not easily computed as we cannot spearate the trend in the data given the assumptions of the model on what constitutes the trend.[/quote]

Thank you very much Johhanes. The third question confused me for a long time and you help me understand it clearly.

[quote=“Massimo”]I’m try to answer to your questions.

I think it’s yes.
With loglinear options, you log-linearize the model around the steady state. So the otputs are percentage deviation from the steady state.

I’m not very familiar with this literature, so I’m not 100% sure that what you say is correct. In particular I’m not sure of the analogy between gross and net rates.

3)I think it depends on what you whant to claim/explain, but if you what to deal with output I think the graph of y is better.[/quote]