This is the teaching repository for the mini-course Machine Learning for Dynamic Economic Models, taught at the Duke University Department of Economics, in September 2026.
Instructor: Yucheng Yang — University of Zurich; ERID Visitor, Duke University · yucheng.yang@uzh.ch
Lecture slides, code, and readings are posted here as the course approaches.
The lectures take place over lunch, 12:00–1:15 PM:
| Lecture | Date | Room |
|---|---|---|
| 1. Deep Learning for Solving Heterogeneous Agents Models | Tuesday, September 8 | SS105 |
| 2. Structural Reinforcement Learning for Macroeconomics | Wednesday, September 9 | SS113 |
| 3. Deep Learning for Continuous Time Models and Structural Estimation | Friday, September 11 | SS113 |
- Reading List (PDF) — suggested preparation plus core and background readings
The course covers three machine-learning methods for solving heterogeneous agent models with aggregate shocks, moving from discrete time to continuous time and from a distribution-based to a price-based state space.
| # | Lecture | Slides | Method | Core reading | Code |
|---|---|---|---|---|---|
| 1 | Deep Learning for Solving Heterogeneous Agents Models | Use neural networks to parameterize high-dimensional value and policy functions in heterogeneous agent models, with the cross-sectional distribution represented by learned generalized moments; trained along simulated paths. | Han, Yang & E (2026), Quantitative Economics | Tutorial 1: DeepHAM on Colab | |
| 2 | Structural Reinforcement Learning for Macroeconomics | Replace the distribution with low-dimensional prices as state variables; agents learn equilibrium price dynamics from simulated paths and optimize via structural policy gradient. | Yang, Wang, Schaab & Moll (2025) | SRL tutorials | |
| 3 | Deep Learning for Continuous Time Models and Structural Estimation | Search and matching with two-sided heterogeneity in continuous time: general equilibrium as a high-dimensional PDE with the distribution as a state variable, solved globally by deep learning and estimated via SMM. | Payne, Rebei & Yang (2026), conditionally accepted, Econometrica | Tutorial 3: DeepSAM on Colab |
Materials: Lectures/ (slides) · Tutorials/ (code walkthroughs) · Readings/ (papers).
The lectures will be easier to follow if you have looked at the two suggested items in the Reading List:
- Heterogeneous-agent models — Dirk Krueger, An Introduction to Macroeconomics with Household Heterogeneity (lecture notes), Chapter 6.
- Coding — Python notebooks on solving simple Brock–Mirman models with neural networks: deterministic and stochastic.
New to Python? See the Python refresher.
If you plan to run the code, a Google Colab account with GPU access is the easiest setup — the tutorial notebooks open in Colab directly and need no local installation.
A longer, five-day treatment of this material and of machine learning for macro-finance more broadly: the PKU–Zurich PhD Summer School on Machine Learning for Macroeconomics and Finance (Beijing, July 2026).
Teaching materials in this repository are released under the Creative Commons CC0 1.0 Universal license, except where individual files state otherwise (e.g. third-party code retained under its original license).