No, you can combine linearized and nonlinear equations provided that i) you only use a first order approximation and ii) correctly link the levels and percentage deviations. If z is the level of a variable, then
zhat=log(z)-log(steady_state(z))
1 Like