Working Papers

The standard approach to discretizing VARs uses tensor grids. However, when the VAR components exhibit significant unconditional correlations or when there are more than a few variables, this approach creates large inefficiencies because some discretized states will be visited with only vanishingly small probability. I propose pruning these low-probability states, thereby constructing an efficient grid. I investigate how much an efficient grid improves accuracy in the context of an AR(2) model and a small-scale New Keynesian model featuring four shocks. In both contexts, the efficient grid vastly increases accuracy.