relative_irf

Would it make sense to allow option relative_irf also in estimation for producing bayesian irfs? Would it be possible to allow options to the new syntax irf_calibration(<options>)

in particular would be fine to have the relative_irf option

irf_calibration(relative_irf)

I would be OK with this, although I don't find the relative_irf option particularly useful. I would prefer a more flexible interface of type we discussed in #115

@rattoma #740 introduced relative_irf to estimation. Is there anything else required for closing this ticket?

if is also available in irf_calibration ?

No. @houtanb This now seems to be a preprocessor issue. What is still needed is allowing the irf_calibration-block to have the input argument relative_irf (mapping to options_.relative_irf) similar to the way the use_calibration option is passed to the estimated_params_init block. @rattoma I have one additional question after looking up this block: it seems the option is specific to the moment computation, but it will generally affect the IRF display of stoch_simul and estimation, even if these IRFs are not restricted. Is this intended?

I am not sure to understand your question. Shouldn't relative_IRF only apply for irf_calibration and be irrelevant for moment computation?

@rattoma You are right that this option will not affect computations. I missphrased. What I meant is the following. Say you have ls2003a from the tests/gsa folder

options_.relative_irf=1;
irf_calibration;
y(1:4), e_ys, [ -50 50]; //[first year response]
@#for ilag in 21:40
R_obs(@{ilag}), e_ys, [0 6]; //[response after 4th year to 10th year]
@#endfor
end;

Thus, you want to restrict the relative IRFs of y and R_obs. But because options_.relative_irf is generally set, it will also affect the saved results in oo_.PosteriorIRF.dsge and the displayed IRF-figures. If you wanted the regular, absolute IRFs, there is no way to trigger them. Put differently, the scope/function of the relative_irf option is larger than just being used in the IRF calibration. Thus, we need to document this clearly.

Right, it will change the behaviour of irfs in all subsequent calls, which can be however reverted? Can we allow a syntax relative_irf=0 and not only relative_irf ?

Maybe a stupid question, but can you have multiple irf_calibration blocks? In this case, we need to worry about conflicting options.

not really for irf_calibration, but to tackle your worry, there can be first irf_calibration

afterwards

stoch_simul

for example, where you would like to plot the levels.

Perhaps, we should set an option for irf_calibration that does not alter the global options_ but only works locally. thus relative_irf must be reset to the value it had before entering calibration. what do you think?

@rattoma The problem you describe is a neverending problem. I would not tackle it as it is common and pretty obvious because the figure titles will show you what happened. I was rather worried whether Dynare automatically concatenates several irf_calibration blocks to one joint block as is the case for example with the estimated_params block. Because in this case, the user could specify conflicting options, trying to have on restriction on relative IRFs and the other on absolute IRFs. If this is not possible, I am satisfied.

So far, only single relative or absolute irf restrictions are allowed, so I think we are OK. thank you