You have to map your stationary model to the data, which describes deviation from a steady state or a BGP. This can either be done by pseudo-“stationarizing”/detreding your data by HP-filtering them. This is often the easiest way to obtain deviations from steady state.
Alternatively, you can map your stationary model to the raw data by means of an observation equation as increasingly common. For this procedure, see the following posts
[Simple model. Unkown error!)
[Measurement equations and steady state growth)