AI for Economic Research: Dynamic Models, Language, and Agents

Zhigang Feng · Professor of Economics

E-mail: z.feng2@gmail.com

Webpage: https://sites.google.com/site/zfeng202/

This course covers what modern AI changes about how economic research is actually done. It runs in six sessions: an orientation to AI as a tool, an economic object and an agent; deep learning, where functions become trainable and equations become losses; reinforcement learning, where decisions are learned from interaction; heterogeneous-agent models, where distributions become state variables; text and large language models as data, measurement and generation; and retrieval and agents in one research workflow. The organizing premise is that as AI absorbs more of the implementation, the economist's edge shifts to designing algorithms and validating results.

Every session ends in a lab: a Jupyter notebook that runs offline on a laptop CPU with fixed seeds — no API key, no GPU. Session 1's last deck, The Workbench, sets up everything the labs need: Python, VS Code, git and GitHub.

Prerequisites and where to start. The course assumes you are comfortable with dynamic macroeconomic models — if you would like that theory firmer first, a recorded offering of Advanced Macro Theory covers it in twenty-eight lectures, from the Solow model through recursive optimization, competitive equilibrium, and asset pricing. Python experience helps but is not required: the companion course, Quantitative Macroeconomics with AI and Machine Learning, teaches it from scratch in Topic 3, Programming Basics for Economists, and covers the classical computational methods — dynamic programming, perturbation, projection, parallel computing. The three are designed as one sequence: macro theory → classical methods → machine learning → agentic AI research, and this site is the last stage.

📋 Syllabus Sessions, decks, resources 📑 Slides Session decks (© Z. Feng) 💻 Labs Notebooks — cloud or local 🎓 Projects Four tracks — pick one

Start here

Three steps per session

Read the session in the syllabus — it lists what each deck covers and what the lab asks.
Download the decks from the slides page — one at a time, or the whole session in one PDF.
Run the lab. The notebooks are the point where the material becomes yours — each runs offline on a laptop CPU.

Where to enter. Everyone starts with Session 1; its last deck, The Workbench, teaches the Python the labs are written in and sets up the tools. For more Python, the companion course's Topic 3 teaches it from scratch. With classical computational training already in hand, Sessions 2–4 (deep learning, reinforcement learning and heterogeneous agents) are the bridge you are looking for. Interested only in text and agents, Sessions 5–6 stand on their own. When you want to build something, go to the project tracks.

Class Meetings — Fall 2026 · 上課安排

Date	Time	Location	Session
Tue, October 13	Morning 9:30–12:30	TBA · 待定	Session 1 — Orientation
Tue, October 13	Afternoon 14:30–17:30	TBA · 待定	Session 2 — Deep learning
Thu, October 15	Morning 9:30–12:30	TBA · 待定	Session 3 — Reinforcement learning
Fri, October 16	Morning 9:30–12:30	TBA · 待定	Session 4 — Heterogeneous agents
Tue, October 20	Morning 9:30–12:30	TBA · 待定	Session 5 — Text and LLMs
Thu, October 22	Morning 9:30–12:30	TBA · 待定	Session 6 — RAG and agents

Course Map · 課程結構

Session	Object	What changes	Decks
1	Orientation	AI as tool, as object, and as an agent with incentives	5
2	Deep learning	Functions become trainable; equations become losses	6
3	Reinforcement learning	Sequential decisions are learned from interaction	6
4	Heterogeneous agents	Distributions and equilibrium become state variables	4
5	Text and LLMs	Language becomes data, measurement, and generation	6
6	RAG and agents	Retrieval, tools, traces, and governance in one workflow	5

Sessions

Session 1 · 5 decks, 140 pages Orientation AI in science · the economics of intelligence · what AI is · workhorse macro models · the workbench Slides Available Lab

Session 2 · 6 decks, 172 pages Deep learning a digit and a neuron · nonlinearity and approximation · backpropagation · PyTorch and a yield forecast · the growth model on its Euler equation · architecture Slides Available Lab

Session 3 · 6 decks, 162 pages Reinforcement learning the growth model with one thing hidden · the simplest actor–critic · Monte Carlo · temporal difference · from tables to networks · across algorithms Slides Available Lab

Session 4 · 4 decks, 168 pages Heterogeneous agents Aiyagari, solved twice · Krusell–Smith three ways · cohorts under aggregate risk · optimal taxation without commitment Slides Available Lab

Session 5 · 6 decks, 208 pages Text and LLMs text as data · embeddings and the Transformer · pretrained models and five limits · the LLM as an instrument · prompts and reproducibility · the yield lab Slides Available Lab

Session 6 · 5 decks, 176 pages RAG and agents what an agent is · how the loop works · building with agents · your own RAG · GRAM Slides Available Lab

Where this course sits

This is the last of three courses that run as one sequence. Advanced Macro Theory supplies the models — twenty-eight recorded lectures from the Solow model through recursive optimization, competitive equilibrium, asset pricing, and fiscal policy. Quantitative Macroeconomics with AI and Machine Learning solves them on a computer — twelve topics on value function iteration, time iteration, EGM, perturbation, projection, and high-performance computing, with slides, labs, and a complete video series (free) from 2026 and 2025, and it is the written subject of the textbook 机器学习与数量宏观经济学. This site takes them to the AI frontier.

Macro Theory → Classical Methods → ML & Deep Learning → Agentic AI Research

You can enter at any stage. This site is the last one.

New here? Start with the full syllabus for the session-by-session plan, prerequisites, and supplementary resources. The earlier ten-lecture version, taught at Zhejiang University in Summer 2026, is kept on the Summer 2026 page.