Ten lectures grouped into six modules of roughly four
hours each — about 24 hours in total. Each module pairs one or two
lectures with a hands-on lab. Taught as an intensive it fits into a week or two.
Level: upper-undergraduate through graduate. All labs run on CPU
alone and work offline once the data is bundled; in practice the deep-learning and
reinforcement-learning labs train noticeably faster on a GPU, so use one if you have access.
Class Meetings — Fall 2026 · 上課安排
Date
Time
Location
Module
Tue, October 13
Morning 9:30–12:30
TBA · 待定
Module 1 — Foundations (Lec 1–2)
Tue, October 13
Afternoon 14:30–17:30
TBA · 待定
Module 2 — ML Essentials (Lec 3)
Thu, October 15
Morning 9:30–12:30
TBA · 待定
Module 3 — Deep Learning for Macro (Lec 4)
Fri, October 16
Morning 9:30–12:30
TBA · 待定
Module 4 — RL & Heterogeneous Agents (Lec 5–6)
Tue, October 20
Morning 9:30–12:30
TBA · 待定
Module 5 — LLMs & Text (Lec 7–8)
Thu, October 22
Morning 9:30–12:30
TBA · 待定
Module 6 — Agentic AI & Case Studies (Lec 9–10)
Course Ecosystem
This course sits at the frontier of a larger body of teaching and research. The complete foundation in classical computational methods — value function iteration, perturbation, projection, EGM, and high-performance computing — is the companion course, Quantitative Macroeconomics with AI and Machine Learning, which has its own slides, labs, and complete video series (free, online) from 2026 and 2025; the same ground is covered in print by the textbook 机器学习与数量宏观经济学 by Zhigang Feng. Together they form a three-stage arc:
Classical Methods→ML & Deep Learning→Agentic AI Research
Students can enter at any stage. Those new to computational economics are encouraged to explore the video series alongside this course; those with classical training will find the bridge to modern AI tools immediately actionable.
Course Description
Economics is being transformed. AI tools that once belonged to computer science labs are now reshaping how economists model complex behavior, extract signal from vast text corpora, and automate the entire research pipeline. This course equips students — at the upper-undergraduate through graduate level — to work at that frontier.
The course runs across six modules: foundations and what AI is; machine-learning essentials for macroeconomists; deep learning for solving dynamic models; reinforcement learning and heterogeneous-agent models; large language models and text as economic data, including retrieval-augmented generation; and agentic AI with case studies. The organizing premise is that as AI absorbs more of the implementation, the economist's edge shifts to designing algorithms and validating results.
Applications span macroeconomics, finance, and economic policy — the text work ranges from FOMC minutes to congressional records, and the agentic tools cover every stage of the research workflow from literature review to replication. The specific topics will evolve as the frontier moves.
The course builds from foundations in dynamic programming and quantitative methods toward the cutting edge of AI research. Students arrive with economic intuition; they leave with the technical fluency and critical judgment to use AI as a genuine research partner.
What You Will Build: Four Integrated Competencies
Theoretical & Algorithmic Mastery — Master economic theory and numerical methods on paper before touching code
Technical Fluency — Working knowledge of Python, PyTorch, and the modern computational ecosystem
AI-Augmented Implementation — Specification-driven development through detailed prompts to AI coding assistants
Critical Validation — Shift from writing code to becoming a rigorous validator of technical correctness and economic plausibility
Prerequisites: Intermediate-level microeconomics and macroeconomics, and comfort with dynamic macroeconomic models. Familiarity with dynamic programming or recursive methods is a plus but not required — the course builds up what it needs. Python experience helps but is not required — the companion course teaches it from scratch in Topic 3, Programming Basics for Economists. No prior ML experience required.
Course Outline
Six modules, ten lectures. Every module ends in a lab; the labs are the part that makes the
material stick. Slide decks are on the slides page, notebooks on
the labs page.
Module 1: Foundations
≈4 hours — Lectures 1 & 2 + environment setup
Lecture 1: Introduction — AI and Economic Research (1 hr)
The state of AI as of early 2026: a rising frontier, falling costs, and historically fast adoption and diffusion
Why economists should care: AI as a general-purpose technology — and what does not qualify as one
AI as cognitive capital: task-level production, the adoption margin and S-curve, intangible non-rival capital and the measurement gap
How AI enters growth: three channels, a light growth-accounting lens, and why task-level gains don't auto-sum to macro TFP
AI as a research collaborator: compressing the theory→code loop and learning from failures (HA-Ramsey illustration)
Context, not capability, is the bottleneck: four ways to give an LLM context and why naive RAG falls short for economics
A map of AI tools for economic research, concrete examples, strengths/limitations, and a task-level AI capability index
Course roadmap: the two arcs (dynamics and language), the 5-pillar “research architect” workflow, reproducibility discipline, tools, and project tracks
Lab — Lec01_02_Lab_Getting_Started.ipynb: set up VS Code,
Python and PyTorch, then write your first Python. Everything downstream assumes this one is done.
Module 2: Machine Learning Essentials
≈4 hours — Lecture 3 + extended lab Lab
Lecture 3: Machine Learning for Quantitative Macroeconomists (2 hr)
Pattern recognition as motivating illustration (handwritten digits)
Lab (2 hr) — Lec03_Lab_ML_Basics.ipynb: a neural network as
flexible regression, then how it approximates any function; predicting the 10-year
Treasury yield with real FRED data; and a build-it-yourself classifier that reads the Fed's words.
Module 3: Deep Learning for Macro Models
≈4 hours — Lecture 4 + extended lab Lab
Lecture 4: Solving Macro Models via Deep Learning & RL (2 hr)
Motivation: curse of dimensionality and the deep learning opportunity
Running example: optimal growth model
Deep learning approach: parameterize value/policy functions with neural networks
Actor-critic algorithm applied to the growth model
Euler equation method + supervised learning
Structured neural networks: ICNNs for convexity preservation
Lab (2 hr) — Lab4A_Dynamic_Models.ipynb: the classical VFI
benchmark, built as ground truth you can trust; Lab4B_Dynamic_Models.ipynb: the same
model solved gridlessly two ways — deep value function iteration and Euler-equation minimization.
Lecture 6: Heterogeneous Agent Models via RL (1.5 hr)
Aiyagari–Bewley–Huggett: setup, SRCE, existence
Krusell–Smith (1998): aggregate uncertainty, moment reduction
DeepHAM and Deep Equilibrium Nets (DEQN)
OLG with aggregate uncertainty and homotopy training
Mean-field games: HJB, KFE, connection to HA models
Topic decks:in preparation
— Aiyagari & computation · why ML + Euler · global DNN · KS & DeepHAM ·
DEQN & OLG · mean-field games (HJB, KFE)
Lab — Lecture 5 — Lab5A_RL_Nutshell.ipynb: reinforcement
learning from tables to neural networks along a MODEL → SAMPLE → SCALE
arc — a gridworld value-iteration benchmark, then Monte Carlo / TD / Q-learning,
ε-greedy vs. Boltzmann, a mini-DQN (replay + target net), and an actor–critic on a
consumption–savings problem, each validated against the exact DP solution. Offline
(NumPy + PyTorch).
Lab — Lecture 6In preparation
— Lab6_KS1998_DEQN_ComingSoon.ipynb: Krusell–Smith (1998) solved with Deep
Equilibrium Nets in PyTorch; calibration, simulation, and Euler-equation accuracy validation.
Module 5: Language Models & Text as Economic Data
≈4 hours — Lectures 7 & 8 + lab Lab
Lecture 7: Large Language Models & Text as Economic Data (2 hr)
Tokenization: characters → subwords → tokens; BPE and WordPiece; why economic jargon is tokenized differently from everyday English
Word embeddings: TF-IDF, Word2Vec, GloVe; the geometry of meaning (analogy arithmetic: king − man + woman ≈ queen); polysemy and context-dependence
Transformer architecture deep dive: self-attention (Q, K, V matrices); scaled dot-product attention; positional encoding; multi-head attention
Full transformer layer: residual connections, layer normalization, MLP sublayer; encoder vs. decoder vs. encoder–decoder
Training paradigms: pretraining on raw text, fine-tuning for downstream tasks (full fine-tuning, LoRA, adapters), and zero-/few-shot prompting of large generative models
Text as economic data — frontier applications: monetary policy stance and inflation forecasting; large-scale policy document classification; LLM embeddings for asset pricing factors and portfolio analysis
Beyond dense retrieval — vector search, hybrid retrieval, and GraphRAG, with a research-grade case study: “wrong-by-prominence” vs. condition-aware retrieval over the economics literature (GRAM)
Labs (1 hr) — four offline notebooks, no GPU, no API key, no network:
Lec07_LLM_Lab/Lab7A_Text_as_Data.ipynb: tokenization and a toy BPE; TF-IDF; a mini
economic-policy-uncertainty index; word embeddings from co-occurrence; a hawk–dove stance score
validated against real rate cycles — where it works and where it fails
Lec07_LLM_Lab/Lab7B_Attention_MiniGPT.ipynb: scaled dot-product attention by hand, then a
minimal GPT trained from scratch on a bundled 30-year FOMC statement corpus (∼10 min on CPU) and
text generation with temperature/top-k
Lec08_RAG_Lab/Lec08_Lab_RAG.ipynb: the full RAG pipeline (chunk → TF-IDF index →
retrieve → audit → cited grounded answer) over a corpus of public speeches, with a minimal
retriever you implement yourself
Lec08_RAG_Lab/Lec08_Lab_MiniGPT_RAG.ipynb — Part B: wire the Lab 7B
mini-GPT to a RAG over the FOMC corpus and compare the answer with and without retrieval, then
measure why a 0.6M-parameter model cannot read its own context
Module 6: Agentic AI & Case Studies
≈4 hours — Lectures 9 & 10 + lab Lab
Lecture 9: Agentic AI for Research Workflows (1 hr)
One map, five acts — what agentic AI is, how it works, how to use it well, and what your role is; two research threads run throughout: HA-Ramsey (Lec 10 T8) and the ES Fellows dataset (Lec 10 T4)
What is agentic AI? Chatbot → RAG → Agent spectrum; the agentic loop and the four features
The agentic-AI landscape (mid-2026): the interoperability stack (MCP, A2A — now under the Linux Foundation) and orchestration patterns (orchestrator–worker, reflection, hand-off)
Reasoning models & inference-time compute; autonomous coding and computer-use agents
Under the hood & how to start: install and initialize, the shape of the CLI, permissions, your first session — Claude Code as the running example
Workflow design: the homotopy workflow (spec first, grade the plan, V0 first, validate then extend) and the OLG five-step demo
The researcher as project manager: goal → final product → intermediate goods → quality gates → assignment to agents and sub-agents → cross-verification; project structure: CLAUDE.md, skills, agents, rules, hooks
The 5-pillar workflow for computational economists
Tracing Aiyagari end-to-end; failure modes; managing context: the context window, compaction, and the Context Dilemma
Case 1: Literature review automation (LLM + RAG on NBER papers)
Case 2: Structured data extraction (FOMC, congressional records)
Case 3: Empty folder to figure (end-to-end empirical workflow)
Case 4: Computational macro model (agentic AI assisting solver/debugger)
Topic decks:in preparation
— framing · paper review · MEPS · ES Fellows · the five-stage collaborator ·
the sprint (setup and execution) · synthesis
LabIn preparation —
Lab9_10_Agentic_CaseStudy_ComingSoon.ipynb: guided replication of a case study
end-to-end with an agentic coding tool.
Supplementary Resources
Foundations: Classical Methods — Companion Course, Video Series & Textbook
This course is part of an integrated curriculum. Three foundational resources cover the classical side of the arc, all freely available:
Companion course — Quantitative Macroeconomics with AI and Machine Learning, the sister site to this one. Twelve topics with slides and labs: numerical methods, dynamic programming and accuracy assessment, perturbation and projection, high-performance computing, and machine learning for heterogeneous-agent and OLG models. Start here if you want the classical foundation this course assumes:
Video Lecture Series — the same course taught on camera: value function iteration, time iteration, endogenous grid method (EGM), perturbation methods, projection methods (Chebyshev), numerical optimization and interpolation, and high-performance computing (MPI, GPU). Nine lectures, about nine and a half hours:
Textbook — 机器学习与数量宏观经济学 by Zhigang Feng bridges classical numerical methods and modern deep learning for dynamic economic models. It is the primary reference for Modules 1–4 and connects to the AI frontier covered in Modules 5–6: