Pruning in DSGE Models - Theoretical Foundations and Comparisons


We study the rationale and performance of DSGE perturbations that are pruned to guarantee stable simulations. We provide theoretical bases for pruning algorithms with matched perturbations of series expansions and nonlinear moving averages. We show that the theoretical rationales they provide differ only in their risk corrections with the series expansion at the stochastic steady state identical to the standard nonlinear moving average. We compare twelve different pruning algorithms, derived here and from the literature, at second and third order, documenting the algorithms in a unified notation. In general, the standard nonlinear moving average or series expansion at the stochastic steady state is the most accurate and the series expansion at the deterministic steady state the second most accurate; yet the theoretically founded algorithms perform comparably, suggesting that this choice is unlikely to be a potential source of error. Alternative ad hoc algorithms from the literature can suffer a loss of accuracy when they include terms inconsistent with or neglect terms consistent with the order of approximation.

Working Paper
Alexander Meyer-Gohde
Alexander Meyer-Gohde
Professor of Financial Markets and Macroeconomics

My research interests include macroeconomics, macro-finance, econometrics, and numerical methods