A Community Reference Library for Computational Economics

Over a dozen toolkits — no shared reference

The field has over a dozen mature toolkits, across MATLAB, Python, and Julia:

Dynare MATLAB / Jl VFI Toolkit MATLAB GDSGE MATLAB RISE MATLAB HetSol MATLAB G-Cubed / SYM Python / SYM HARK / Econ-ARK Python GEModelTools Python sequence-jacobian Python Mesa Python Dolo / Dolark Py · Jl BASEforHANK Julia Eurace / FLAME C++ AGENTBLOCKS

Each has genuinely distinct strengths. But researchers have no easy way to compare them, no shared library of canonical models, and no on-ramp to the ecosystem as a whole.

Real diversity in how we solve models — and no easy way to see across it.

A Community Reference Library

A single, community-owned library documenting the field’s canonical models — the same model for every toolkit family, side by side, open and runnable.

Community-owned

Authored & maintained by the projects themselves — each keeps credit and editorial control.

Cross-toolkit

The same canonical models across MATLAB, Python & Julia — comparable, side by side.

Open & executable

Every model runs — open-source and reproducible, published on Jupyter Book.

The shared library — and the on-ramp — the field has been missing.

The proposal

Each project authors its own work and contributes two things:

1

Shared baseline modelcoordinated across your group — directly comparable across toolkits

2

Hero tutorialyour author's choice — the problem where your toolkit shines

Family	Shared baselines	Toolkits coordinated around them
Heterogeneous-agent	Aiyagari · 1-asset HANK	VFI ToolkitHetSolHARKDoloGEModelToolsSSJGDSGEBASEforHANKDynare v7
Representative-agent / DSGE / CGE	RBC · 3-eq NK (RANK)	DynareDoloSSJGDSGERISEG-CubedDynare.jl
Agent-based	Boltzmann–Gibbs · Lengnick ABM	MesaEurace / FLAMEAGENTBLOCKS

A hero tutorial enables the author to showcase the toolkit — for example, G-Cubed could choose a multi-region CGE model as a unique problem this package can solve. The baseline models proposed for each family are a starting point and open to discussion.

Here’s the proposal, made concrete. A community-owned library — same model, multiple tools, side by side. Crucially, each project authors and maintains its own implementations. Three families where several toolkits genuinely overlap; for each, two shared baselines — an introductory rung and an advanced rung — so a group can grow from the simple case into the hard one. We coordinate the projects in that row so a reader can open Aiyagari in VFI Toolkit next to Aiyagari in HARK next to Aiyagari in SSJ — same notation, directly comparable — and then step up to the 1-asset HANK baseline the same way. And each project also contributes an author’s-choice hero showcase for the problem where its toolkit shines. G-Cubed, for example, could choose a multi-region CGE model — a unique problem this package can solve that none of the shared baselines reach, which is where CGE fits in this proposal: as a showcase, not a baseline. Exact coverage is for the working group to set; this is the shape of it.

What a “shared baseline” means — one idea

Representative agentDSGE

Heterogeneous agentIncomplete markets

Agent-basedBottom-up macro

Introductory

RBC growth

Real business cycle — the DSGE starting point

Introductory

Aiyagari

General equilibrium — wealth as an equilibrium object

Introductory

Boltzmann–Gibbs exchange

An emergent distribution → Gini

Inequality →equilibrium vs emergent — one question, two mechanisms

Advanced

3-equation New Keynesian

RANK — representative-agent closure

+ 25bp monetary shock

Advanced

1-asset HANK

Heterogeneous-agent New Keynesian

+ 25bp monetary shock

Advanced

Lengnick (2013)

Baseline macro–ABM

Same 25bp shock →overlay the two IRFs — the signature RANK vs HANK comparison of the last decade

Input Shared Specification the agreed standard

→

What we decide next Decisions

1Income process & grids
2Model closure
3Calibration targets
4Comparison contract

→

Output Jupyter Notebooks with a model to solve

This is what “shared baseline” means in practice. Three families across the columns, two rungs down each — an introductory model and an advanced one — so a group can start simple and grow into the hard case. Representative-agent: RBC up to a three-equation New Keynesian (RANK). Heterogeneous-agent: Aiyagari up to one-asset HANK. Agent-based: a Boltzmann–Gibbs money-exchange model up to Lengnick’s baseline macro-ABM. The two coloured bands are the punchline. First: the same 25-basis-point monetary shock runs in both advanced rungs — RANK and HANK — so overlaying the two impulse responses gives the signature RANK-versus-HANK comparison of the last decade. Second: inequality links the two introductory rungs — in Aiyagari the distribution is an equilibrium object, in the agent-based model it emerges bottom-up — one question, two mechanisms. A word on comparability, because this is where it gets real: pinning the income process and grids matters, but so does the model closure — the Taylor rule, the shock, and especially the fiscal rule, since Ricardian equivalence fails in HANK — and even then, solution-method differences move results. So the contract differs by family: DSGE and HANK compare on impulse responses and steady states, the heterogeneous-agent rungs add distributional moments, and agent-based models compare on ensemble moments across seeds, not a single path. One thing to flag if pressed: a one-asset HANK calibrated to aggregate wealth has low MPCs and looks close to RANK, so we’d calibrate to a target MPC for the comparison to show what it should. Rather than hand any of this down, we’d agree the spec with the working group and distribute it as template Jupyter notebooks — defaults like Rouwenhorst with seven states are a starting point, not a decree.

What the proposal aims for

A community asset

The field's first cross-toolkit reference — built by and for the community. A single entry point that lowers the search cost of a fragmented field, lets researchers compare toolkits side by side, and showcases each project through its hero tutorial — all on Jupyter Book, a modern code-first publishing platform.

Building towards

Substrate for AI assistance

Exactly the corpus an AI research assistant needs:

"Which tool fits my incomplete-markets model?" "Translate my Dynare model to sequence-jacobian."

A proving ground for standards

As shared schemas emerge — steady states, IRFs, Jacobians — the library demonstrates them in working code, and becomes the testbed for Econ-ARK's Project Bellman semantic representation.

One priority, and two things it builds toward. The priority — first and foremost — is a community asset: the field’s first cross-toolkit reference, built by and for the community. It’s a single entry point that lowers the search cost of a fragmented field — over a dozen toolkits with no easy way to compare them; it lets researchers put toolkits side by side on the same models; and it gives each project a showcase through its hero tutorial. And it’s all published on Jupyter Book, the modern code-first platform we already run at scale. That’s the foundation, and it delivers value on its own. From that foundation, two things follow. One — the part I’m most excited about — it becomes the substrate for AI assistance. An assistant for this field is only as good as the corpus behind it, and right now that corpus doesn’t exist in structured, cross-toolkit form. This library is that corpus: it lets an assistant answer “which tool fits my incomplete-markets model?” or “translate my Dynare model to sequence-jacobian” with grounded examples — and it’s exactly what Econ-ARK’s semantic-representation work needs to validate against. And two, it becomes a proving ground for the standards this working group will define, demonstrated in working code, not just specified on paper. But the foundation comes first — the library itself. QuantEcon brings the engineering capacity to build the glue — which I’ll come to.

QuantEcon Infrastructure

A community partner — already running open-source executable book infrastructure:

~31kmonthly users

172countries

274lectures

Executable Jupyter Books

Code-first publishing — prose, maths, and runnable code in one source that executes and renders straight to a website. MyST / JupyterBook 2, the stack behind our lecture sites, with a book theme we maintain.

Multi-language execution & CI

Python & Julia in Jupyter — with or without GPU instances for compute-intensive workflows. Interactive tutorials via Binder or Colab.

MATLAB support

Run locally via the MATLAB kernel for Jupyter — MATLAB needs a licence, so we publish pre-rendered output. Home for Dynare, VFI Toolkit, GDSGE.

The ecosystem glue

A GitHub-based auto-publishing platform, hosting for community pages, and help with continuous integration and testing.

A two-phase roadmap

Phase 1end of 2026

Community library

A static Jupyter Book site publishing tutorials on:

2 × baseline models — for cross-comparison of toolkits
1 × hero tutorial — per package / toolkit
CI runs to check execution and report execution issues

Phase 22027+

Reference standard

A library to test shared schemas
Cross-toolkit validation checks
Semantic examples from Econ-ARK Project Bellman — semantics & schema

The ask — and who does what

Register to participate today — plus, for Phase 1 (by end of 2026):

Shared baseline models — introductory and advanced, coordinated across your group — and a hero tutorial, your author’s choice: the problem where your toolkit shines
Authored and maintained by your project — you keep credit, editorial control, and the right to update or withdraw

Projects

Author & own MyST content (Markdown or another format if preferred)
Keep credit & editorial control
Collaborate via GitHub

QuantEcon

Hosts, publishes & builds the glue
Develops supporting infrastructure (e.g. MATLAB)
A shared home that champions every project

Governance sits with the SCE working group launched at this meeting.

Same model. Multiple tools. Side by side.

The field’s first cross-toolkit reference — built by the community that owns these tools.

Register to participate — today
Shared baselines + a showcase from your toolkit — by end of 2026

Register your project → github.com/QuantEcon/conference-cef2026 · Let’s talk after the session.

mmcky@quantecon.org

Support slides

Reference material — for questions and discussion.

Who is QuantEcon?

A non-profit building open-source tools for computational economics — used across research, central banks, and universities worldwide.

~31kmonthly users

172countries

8lecture series

30+workshops delivered

Lectures

Open-source curriculum from intro to advanced — Python, Julia & JAX. Free to read, run, and adapt.

Books

Peer-reviewed textbooks — Dynamic Programming, Economic Networks, Quantitative Macro Finance.

Infrastructure

The executable-book stack — MyST & JupyterBook 2, plus the QuantEcon.py / .jl libraries. Founding member of the Executable Books Project.

Training

Short courses delivered at the IMF, central banks, and universities worldwide.

Reach: quantecon.org analytics dashboard — monthly active users averaged over Dec 2025–May 2026, across 172 countries.