Observable unit root

Hi Prof,

I am a bit confused regarding observables (raw data) that are non-stationary in stationary versus non-stationary models.

Case 1. When the model bloc has a unit root equation such as the definition of inflation in trend and gap model:

\pi_t = p_t - p_{t-1}

where p_t = log(P_t)

Then using p_t as observable (despite being unit root process) is safe as both the observable and the model variables matches. The only thing that we have to take care of is to override the steady state computation using any arbitrary value since the model is a linear model. (We have already discussed this in a previous post)

Case 2: If the model is a NK model which is entered in dynare after log-linearization and no unit root equation is present in the model bloc but an observable (price per se) exhibits a stochastic trend. In this case, **if I add an auxiliary equation \pi_t = p_t - p_{t-1} and use the observable p_t would suffice ** ?

This is because I have seen in some posts that some users are adding dp{obs_t} = p_t - p_{t-1} and using the demeaned log difference of the price level P (i.e: stripping the trend from dp_obs ) and I think these two different ways of dealing with a non-stationary observable of case 2 are equivalent . Do you agree ?

Sorry for this basic question but wanted to double check.

No, those two are not equivalent. A growth rate of prices will not contain any information on the initial price level, while observing prices will have that.