Fast Upper-Envelope Scan for Discrete-Continuous Dynamic Programming
42 Pages Posted: 12 Aug 2022 Last revised: 22 Oct 2024
Date Written: August 4, 2022
Abstract
We develop a general fast upper envelope scan method to solve any stochastic
dynamic models with discrete-continuous choices where the optimization problem
has a one-dimensional representation. For this vast class of models, FUES enables
the endogenous grid method by efficiently selecting the optimal solution using a targeted
secant scan. Importantly, it does so without requiring either monotonicity of the
policy function or analytical information on the value function gradient. We demonstrate
our method using several workhorse applications in the literature, and show
the massive computational speed and accuracy gains generated across the board, even
when uncertainty may lead to highly complex endogenous grids of potential solution
points.
Keywords: discrete and continuous choices, non-convex optimization, Euler equation, dynamic programming
JEL Classification: C13, C63, D91
Suggested Citation: Suggested Citation