11-02, 10:55–11:35 (America/New_York), Winter Garden (Room 5412)
Economics research has widespread policy and industrial applications. It is rapidly changing due to open source tools. Deep learning techniques have widened the range of models that it is feasible to solve, simulate, and estimate. This talk highlights recent contributions to computational economics and their Python packages. It is aimed at quantitative analysts and economists working in finance and public policy.
This talk discusses new open source Python tools used for computational economics research. It describes tools and techniques at a high, general level, with some mathematics.
The talk begins with a discussion of general economics tools for heterogeneous agent modeling: QuantEcon, Dolo, and HARK.The core problem faced in computational economics research is that until recently, dynamic choice problems with large state spaces were computationally infeasible to solve. I will discuss the main pain points with heterogeneous agent modeling, namely: solving dynamic choice problems in which agents act in high-dimensional state spaces, and estimating models with many free parameters.
I will then discuss highlights of recent research that uses deep learning to overcome these computational limits. Deep Equilibrium Net techniques can solve high-dimensional models by building first order conditions directly into the neural network loss function. Deep estimation techniques allow for efficient estimation by including free parameters as additional quasi-state in network training.
I will conclude by advocating for unifying these techniques with an interoperable modeling API.
- 2 minutes: Intro
- 2 minutes: QuantEcon: Introductory training materials in computational economics.
- 3 minutes: HARK. A framework for heterogeneous agent models.
- 5 minutes: Dolo.py and dolang: A language and system for defining and solving economic models.
- 8 minutes: The computational challenges of heterogeneous agent modeling.
- 10 minutes: Deep Equilibrium Net methods for solving high-dimensional problems.
- 5 minutes: Deep estimation techniques.
- 5 minutes: Call to action. A unified API for interoperating between these techniques.
Links and references:
https://econ-ark.org
https://github.com/econ-ark/HARK
https://www.econforge.org/dolo.py/
https://github.com/EconForge/dolo.py
Azinovic, Marlon and Gaegauf, Luca and Scheidegger, Simon, Deep Equilibrium Nets (May 24, 2019). Available at SSRN: https://ssrn.com/abstract=3393482 or http://dx.doi.org/10.2139/ssrn.3393482
https://github.com/sischei/DeepEquilibriumNets
Chassot, J., & Creel, M. (2023). Constructing Efficient Simulated Moments Using Temporal Convolutional Networks.
https://www.jldc.ch/publication/2023_deep-simulated-moments/
https://github.com/JLDC/DeepSimulatedMoments.jl
Previous knowledge expected
Sebastian Benthall, PhD, is a contributor to Econ-ARK, an open source toolkit for heterogeneous agent modeling. He is a Principal Investigator at the International Computer Science Institute, and a Senior Research Fellow at New York University School of Law, where he researchers computational economics approaches to data protection and AI regulation.