The interaction of macroeconomic variables may change as the nominal short-term interest rates approach zero. In this paper, we propose an empirical model that captures these changing dynamics with a time-varying parameter vector autoregressive process. State-dependent parameters are determined by a latent state indicator. This state indicator follows a distribution with time-varying probabilities affected by the lagged interest rate. As the interest rate enters the critical zero lower bound (ZLB) region, the dynamics between the variables and the effect of shocks change. We estimate the model with Bayesian methods and explicitly consider that the interest rate may be constrained in the ZLB region. We provide an estimate of the latent rate, i.e., a lower interest rate than the observed level, which is state- and model-consistent. The endogenous specification of the state indicator permits dynamic forecasts of the state and system variables. In the application of the model to the Swiss data, we evaluate state-dependent impulse responses to a risk premium shock that is identified with sign restrictions. Additionally, we discuss scenario-based forecasts and evaluate the probability of the system exiting the ZLB region that is only based on the inherent dynamics.