Numerical Dynamic Programming on Parallel Computing Clusters: A 'MacGyver' Approach

15 Pages Posted: 8 May 2008

See all articles by Florian Zainhofer

Florian Zainhofer

University of Fribourg (Switzerland) - Chair for Financial Management

Date Written: May 2008

Abstract

Economists who want to numerically approximate an elaborate dynamic stochastic programming problem (DSPP), either for structural estimation or policy evaluation purposes, are often confined by the curse of dimensionality: richer models with various state and control variables cannot be solved on standard pc's in reasonable time. Parallel computation offers a back door to escape from this dilemma (see e.g. Rust (1996). In this note, I illustrate how such a parallel computation can be implemented by applied economists using simple household items. Specifically, I present a simple Mathematica code to approximate a canonical DSPP on a parallel computing cluster. My code achieves a significant reduction in computation time relative to the standard serial solution, is very intuitive and can easily be extended to the most complicated problems.

Keywords: Numerical dynamic stochastic programming, computational methods, parallel computing

JEL Classification: D91, C61, C63

Suggested Citation

Zainhofer, Florian, Numerical Dynamic Programming on Parallel Computing Clusters: A 'MacGyver' Approach (May 2008). Available at SSRN: https://ssrn.com/abstract=1032550 or http://dx.doi.org/10.2139/ssrn.1032550

Florian Zainhofer (Contact Author)

University of Fribourg (Switzerland) - Chair for Financial Management ( email )

Fribourg, CH 1700
Switzerland

HOME PAGE: http://www.unifr.ch/finanzmanagement

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
209
Abstract Views
1,351
Rank
274,785
PlumX Metrics