Estimate Responsibly

Reliable Estimation of Structural Economic Models

Philipp Eisenhauer

Scientically well-posed models make explicit the assump-

tions used by analysts regarding preferences, technology,

and information available to agents, the constraints under

which they operate, and the rules of interaction among

agents in market and social settings. (Heckman, 2008)

StructuralEconometrics

Programmer

Economist

Econometrician

Economist

Agent Behavior

I Objectives

I Preferences

I Constraints

I Institutional

I Information

Economic Environment

I Equilibrium

I Shocks

I Interactions

Michael P. Keane, Petra E. Todd, and Kenneth I. Wolpin (2011).

The Structural Estimation of Behavioral Models: Discrete Choice

Dynamic Programming Methods and Applications. Handbook of

Labor Economics, Elsevier.

Agent Decision Problem

Vt(Ωit) = maxdit

E

[τ=T∑τ=t

δτ−t [U1

iτdiτ + U0

iτ (1− d1τ )]

∣∣∣∣∣Ωiτ

]

Uit = cit + α(1− dit) with αit = κitβκ + βnnit + εit

cit = yit + witdit − (π − τ)nitdit

Labor Market

wit = zitγ1 + γ2hit + ηit ,

where hit =∑τ=t−1

τ=1diτ .

Shock Structure

εitηit

∼ N

0

0

,

σ2ε σε,η

σε,η σ2η

Econometrician

Conguration Files

structToolbox-estimation structToolbox-simulation

Conguration Files

I Initialization

I Optimization

Exploring the Commands

@heracles:$ structToolbox-simulation -h

@heracles:$ structToolbox-simulation -init init.ini

@heracles:$ structToolbox-estimation -h

@heracles:$ structToolbox-estimation -init init.ini

@heracles:$ structToolbox-cleanup

Programmer

Economy

Agents Parameters

Decision

Tree

Tree Class

Class that captures the institutional constraints.

Methods

I addNode

I getNodes

Attributes

I nodes

Parameter Class

Class for the parametrization of the economic environment.

Methods

I addParameters

I getParameters

Attributes

I paraObjs

Agent Class

Class that describes the agent behavior in the economic

environment.

Methods

I step

I _evaluateNodes

Attributes

I attributes

I parasObj

I treeObj

Economy Class

Class that collects all economy wide information.

Methods

I simulate

Attributes

I attributes

I parasObj

I treeObj

Additional Topics

I High Performance Computing

Replacement of selected Python functions by Fortran counter-

parts and use of mpi4py for parallel evaluation of likelihood

function.

I Automatic Build Process

Cleanup and compilation of Fortan routines using waf.

I Unit Testing

Automatic execution of unit testing routines during build process

using nosetest.

Final Remarks

Test of Reliability

Fit structural model to

observed data.ψobs

Simulate synthetic dataset

using estimates.

Fit structural model to

synthetic data.ψsyn

Compare

Transparency, Recomputability, and Extendibility

http://www.policy-lab.org/teaching/struct-toolbox

I Documentation

I Source Code

I Test Suite

I Download

Philipp Eisenhauer

Mail eisenhauer@policy-lab.org

Web http://www.policy-lab.org/peisenha

Repository https://github.com/peisenha

Appendix

References

Heckman, J. J. (2008). Econometric Causality. International

Statistical Review, 76(1):127.

Keane, M. P., Todd, P. E., and Wolpin, K. I. (2011). The

Structural Estimation of Behavioral Models: Discrete Choice

Dynamic Programming Methods and Applications. In

Ashenfelter, O. and Card, D., editors, Hanbook of Labor

Economics, volume 4A, pages 331461. Elsevier Science.