Why
Strong toolkits, siloed communities
Computational economics has more than a dozen serious open-source toolkits across MATLAB, Python and Julia, each with distinct strengths — and no easy way to compare them, no shared library of canonical models, no on-ramp to the ecosystem as a whole. Users stay with the first tool they learn, and the projects themselves have no standing channel connecting them.
Community first, corpus second
The first thing this project builds is the community: the toolkit projects united in one group, with common pathways of communication and a shared library of tooling maintained together. The corpus of baseline models is the goal, but it is the community that produces it and keeps it alive.
- A community resource, built together
- The field's first cross-toolkit reference — an on-ramp for new users and a showcase for the projects behind the tools.
- A test corpus for the semantics track
- When toolkit teams state formally what their model files mean, a curated set of worked models is what those statements have to account for.
- Substrate for AI assistance
- Exactly the corpus an assistant needs to answer which tool fits my problem? and translate my model from X to Y.
- A proving ground for standards
- Shared schemas — steady states, IRFs, Jacobians — demonstrated in working code rather than prose.
The library
The working group publishes a set of baseline models. Each project chooses as many as its toolkit can solve — several or just one — and submits a tutorial for each. Buffer-stock saving comes first; four more are proposed.
Buffer-stock saving
Consumption–saving under income risk
Aiyagari · Krusell–Smith
Incomplete-markets heterogeneous agentsA two-asset HANK model
Heterogeneous-agent New KeynesianA small New Keynesian DSGE model
Representative-agent DSGEA baseline agent-based macro model
Macro from interacting heterogeneous agents
Each model description states the shared assumptions and calibration. Each project chooses how to present its implementation. A filled slot names the solution's toolkit, method and submitter, and shows the description version it follows and its AI-assistance category: No AI, AI-edited, AI-assisted or AI-generated; a Verified badge means a person has checked it end to end against the current version.
How it works
Each participating project makes two contributions. The first makes the toolkits comparable; the second lets each project show what it is actually best at.
Baselines from the shared set
Each project solves its chosen baseline models at the shared calibration, reports the results the model description lists, and explains its numerical approach in its own words.
An author's-choice showcase
A tutorial of the project's own choosing, on the problems its toolkit was built for. A common baseline alone would flatten the differences that matter; this is where each project argues its own case.
Both tutorials use the toolkit's own notation and include the authors, software versions and numerical settings needed to reproduce the results. Authors explain their method and any differences from the shared model.
Model descriptions and implementations
Shared model descriptions belong in models/<model>/, with theory, calibration and numerical solution. Runnable solutions belong in solutions/<model>/, each in a directory numbered in submission order, such as 001-hark-egm; each submitter chooses the prose, method and format, such as a notebook or MyST document. Read the buffer stock description and run the HARK example.
Read the model
The model description states the equations, common calibration and numerical solution.
Write a runnable implementation
Keep the model and calibration fixed while choosing the method, prose and format.
Report and check the results
Report the results the model description lists, in any format, and explain the numerical checks used to assess the solution. Optional comparison tables are provided with the HARK example.
Submit your tutorial
Add your runnable files, instructions and metadata.yml in a numbered directory under solutions/<model>/.
- The model specifies functional forms, timing and normalisation.
- The calibration states parameter values and their sources.
- The income specification states the distribution and required moments.
- The model description lists the results each implementation reports.
- Choose the solution method and algorithm.
- Choose the grids or other function approximations.
- Choose how to approximate the shock distributions.
- Use the toolkit's language and notation.
- Choose the prose and runnable format.
Every implementation reports the results the model description lists. The submission guide offers optional formats for result tables and metadata, together with draft comparison tolerances; projects can use them or present the results in another form. Explain any part of the stated model that the toolkit cannot represent. Phase 1 publishes no timings and no ranking of toolkits.
The ask
What each project provides
- Agreement in principle to take part.
- By end of 2026, at least one baseline model solved using the shared description and calibration, so results line up side by side.
- An author's-choice showcase tutorial highlighting where the toolkit is strongest.
- Both contributions authored and maintained by the project — CC-BY for text, an OSI-approved licence for code.
What each project keeps
- Authorship credit for its contributions.
- Editorial control over how its tool is presented.
- The right to update or withdraw contributions at any time.
QuantEcon curates and publishes; it does not own anyone's work or rank the toolkits.
Who does what
QuantEcon
Publishes and maintains the library, sets editorial standards for what goes in, and develops shared tooling in consultation with the projects, as part of QuantEcon's open-source infrastructure.
Projects and contributors
Toolkit projects and individual contributors author and maintain their own solutions and tutorials. A solution stays theirs, and changes to it need its maintainers' approval.
SCE Working Group 1
Builds and releases the baseline set and its model descriptions. The library is run in collaboration with the group; QuantEcon attends as a member to see where it can assist the group's semantics work.
Project A
The Semantics of Economic Models — the working group's companion track, in which toolkit maintainers write a solver-independent account of what their model files mean, published as a journal special issue. The library's worked models are the corpus those accounts have to account for.
Roadmap
Phase 1 · Community library
6–12 mo- 3–4 canonical models in 2–3 toolkits each, published on a static JupyterBook site.
- Open-source repository, PRs welcome.
- Per-project tutorials and per-model metadata.
- CI runs the implementations and flags breakage.
Phase 2 · Reference standard
12–24 mo- Shared output schemas and a queryable index.
- Cross-toolkit validation reports: where do the same model's outputs diverge?
- Library models expressed in a shared semantic representation, building on Econ-ARK's DP schema.
- Documentation of that design, plus a how-to tutorial.
- Exploring an AI advisor built on the corpus.
Presentations
Talks introducing the library, newest first.
Contribute
Start with the buffer stock submission guide. Keep the model and calibration fixed, submit a runnable implementation in your chosen format, and report the results the model description lists. Document the numerical settings and commands needed to reproduce the results, and include a short metadata.yml recording the authors, the description version followed and the AI disclosure, as the contributing guide explains; the output format is your choice.
Anyone can submit a pull request, on behalf of a toolkit's project or as an individual. Add your files under solutions/<model>/, in a directory named <NNN>-<toolkit>-<method> with the next free number; the HARK buffer-stock example is solutions/buffer-stock/001-hark-egm/. Changes to an existing solution need its maintainers' approval.