Hi Michael, Usually the no ponzi game condition is used to select a steady state. For instance, in the Ramsey optimal growth model we have three steady states: (kstar = 0, cstar=0), (kstar>0, cstar=0) and (kstar>0,cstar>0). We do not consider the first one because in this case the economy does not exist. All the trajectories leading to the second one violate the transversality conditon, so finally we select the third steady state as the “good one”. You have to choose initial conditions close enough to the third steady state, to be sure that Dynare approximate the model in the neighborhood of the economically sensible steady state. But the best approach is to define the steady state in a file as described in this post. So that you will be sure that dynare always select the steady state satisfying the no ponzi game condition. If you want to estimate a model, it is always better to define (at least partially) the steady state.
Best, Stéphane.