Are Nonlinear Methods Necessary at the Zero Lower Bound?
27 Pages Posted: 9 Aug 2016 Last revised: 21 Feb 2019
Date Written: August, 2016
This paper examines the importance of the zero lower bound (ZLB) constraint on the nominal interest rate by estimating three variants of a small-scale New Keynesian model: (1) a nonlinear model with an occassionally binding ZLB constraint; (2) a constrained linear model, which imposes the constraint in the filter but not the solution; and (3) an unconstrained linear model, which never imposes the constraint. The posterior distributions are similar, but important differences arise in their predictions at the ZLB. The nonlinear model fits the data better at the ZLB and primarily attributes the ZLB to a reduction in household demand due to discount factor shocks. In the linear models, the ZLB is due to large contractionary monetary policy shocks, which is at odds with the Fed?s expansionary policy during the Great Recession. Posterior predictive analysis shows the nonlinear model is partially able to account for the increase in output volatility and the negative skewness in output and inflation that occurred during the ZLB period, whereas the linear models predict almost no changes in those statistics. We also compare the results from our nonlinear model to the quasi-linear solution based on OccBin. The quasi-linear model fits the data better than the linear models, but it still generate too little volatility at the ZLB and predicts that a large policy shock caused the ZLB to bind in 2008Q4.
Keywords: Bayesian estimation, model comparisons, zero lower bound, particle filter
JEL Classification: C11, E43, E58
Suggested Citation: Suggested Citation