Coarse-graining and model-reduction techniques are widely used to extend the accessible time scales of molecular and stochastic simulations by reducing the number of degrees of freedom. In molecular dynamics, classical coarse-graining approaches typically construct effective Markovian models by approximating the potential of mean force. In contrast, we derive an improved Markovian model through a hierarchical construction of force corrections combined with the estimation of local spatiotemporal rescaling rules. We prove that this new reduced model enhances classical approaches by accurately capturing equilibrium statistics while eliminating systematic errors in dynamical observables, such as the mean-squared displacement. We then extend this framework to general overdamped Langevin systems featuring multiple local minima and bifurcation structures in transition regions, which give rise to metastability and rare-event dynamics. To capture the statistics of transitional behavior, we employ an effective Markovian jump process between metastable states. We show that the resulting reduced models not only reproduce equilibrium statistics, but also accurately predict transition rates and inter-well dynamics. Overall, our analysis demonstrates that the proposed reduced Markovian jump formulation provides a principled and quantitatively accurate description of both equilibrium and dynamical behavior across metastable regimes.