BCA - normalising data

General DSGE Modeling
1

In business cycle accounting, the simulated data (counterfactuals) can deviate from observed data quite strongly.

ewafull

Is that the reason why researchers usually normalize simulated and observed data, for example, to 100, for the start of a given episode? For example, 1979 = 100. I understand that we can normalize if we are concerned about studying short-run dynamics.

aew

My question: So like the big gap between simulated and actual data in the first plot (in the 2nd half of the sample) is sort of meaningless, yeah? Thus, because we have linearly detrended the data, by assumption, that gap should not exist in theory, and it has no meaning, I guess. Or I am wrong? Because no one has tried to explain it in the BCA literature.

And I guess it is also true for all counterfactual analyses that compare simulated data to actual data, yeah? For example, using simult_().

2

There are two different issues here:

  1. The normalization is usually fine, because the model will be in percentage deviations and therefore unitless. Any arbitrary scale can be accommodated as the model reproduces deviations from the mean.
  2. It is BUSINESS CYCLE accounting. The trend is beyond the scope of the exercise. The trend may have an important meaning, it’s just not the job of the model at hand to explain this trend.