Which moments to match (big ratios, standard deviations, correlations, cross-correlations, autocovariance). It seems it is discretionary and depends on the researcher.
Is matching just big ratios ok for IRF analysis? Looking for reference on which moments to match for a particular task.
In the end, it’s all about assigning parameter values for your model. As long as you can justify your values, you should be fine.
Just looking at big ratios will generally not be sufficient as they only inform you about the long-run properties of your model. They will not provide information on the stochastic processes. But for IRFs you need the latter.
Oh I see. Thanks! But like you said in summer school, some referees may say why don’t you match this or that moment although it may not affect the key results you wanna show. So I thought maybe there is some general knowledge out there about which moments to match for which task.
It’s not so much about tasks. Referees want your model to fit the data. If they have the feeling the model does not fit, then nothing you do later on will satisfy them. Unfortunately, it can be very idiosyncratic what they consider to be a good fit.