Optimal Macroeconomic Policy in Nonlinear Models: A VSTAR Perspective

Abstract The paper describes a method to transform vector smooth transition autoregressions in a form that is particularly suitable for policy analysis, because it is of low-dimension and retains certainty-equivalence. Optimal rules are calculated with the state dependent coefficients approach, which allows linear methods to solve nonlinear problems. The methodology is applied to revisit interest rate and quantitative easing (QE) monetary policy in the United States during 1979–2018. The actual size of QE and duration of the zero lower bound are found close to those prescribed by the optimal policy, but not the timing and composition of QE. The methodology compares favourably against alternative optimization approaches based on linearization or numerical search.