Data used in estimation

Hi all,

I would like to ask a question on the estimation of DSGE model. When we estimate the model in log-linearized form, is it necessary to express all data in percentage terms? In other words, shall we use data in the form of:

log(X_t)
or
100*log(X_t)

Will there be any difference in results given by the two?

Thank you so much.

Pam

Hi Pam, This should not make a difference as long as the provided variables in the data file are consistent with the variables declared in the model block. For instance, if you want to use percentage growth rate of GDP you should have something like :

dY = 100*log(GDP/GDP(-1)*ProductionEfficiencyGrowth) ;

in the model block. Where GDP is the detrended GDP and ProductionEfficiencyGrowth is the exogenous productivity growth.

Best, Stéphane.

Dear Stéphane,

Sorry to get back again to this post after such a long time…So if we log-linearized the model by hand and then use data of the form log(X_t), then there will be no problem? I am asking this because it seems some papers scaled the observed variables by 100, which means to express the variables in percentage terms…

Many thanks,

pam

You should also make sure that the priors you declare for the standard deviation of shocks are compatible with your data.