Existing Models and What They Teach Us

Overview

This project does not start from a blank slate. The relevant comparison set already exists: DynaSim, MINT, CBOLT, PWBM, and a small number of open or partially open alternatives. The right question is not “how do we beat them on every dimension?” The right question is “what do they optimize for, what do they leave inaccessible, and where is there still room for a public model to matter?”

Four dimensions are especially important:

  1. Credibility: how much trust the model has earned with agencies and policy professionals
  2. Transparency: how much of the methodology, code, and data logic can be inspected publicly
  3. Scope: what policies and populations the model can analyze
  4. Accessibility: who can actually use the model in practice

DynaSim: the benchmark for non-governmental breadth

Urban Institute’s DynaSim is the closest thing to the benchmark non-governmental dynamic microsimulation model for retirement policy analysis (Favreault et al. 2015).

What DynaSim gets right

  • It has decades of accumulated refinement and institutional memory.
  • It treats Social Security as part of a broader retirement-income system rather than in isolation.
  • It supports rich demographic and economic heterogeneity, including marriage, disability, and asset dynamics.
  • It has produced a large body of policy analysis that researchers and funders recognize.

What DynaSim does not solve for the public

  • The code is not open.
  • Access typically depends on contracts or institutional relationships.
  • Users cannot easily inspect or modify the full modeling pipeline.
  • The model is influential, but not independently reproducible in the way an open research tool should be.

What this project should learn from DynaSim

The lesson is not to imitate DynaSim’s institutional model. The lesson is to match its seriousness about validation, demographic detail, and cross-program context while choosing different tradeoffs on access and transparency. The longer public-source dossier lives in appendix-dynasim.md so this chapter can stay focused on strategic comparison. The more operational side-by-side comparison lives in benchmark-model-component-matrix.md.

MINT: the benchmark for administrative-data credibility

SSA’s Modeling Income in the Near Term (MINT) is the strongest reference point for credibility in Social Security microsimulation (Smith et al. 2010; Social Security Administration 2024).

Why MINT matters

MINT’s defining advantage is access to administrative earnings records. For older cohorts, that removes much of the uncertainty that public-data models have to manage. It also gives SSA a natural validation loop: administrative data continue to arrive, and projection quality can be checked against realized outcomes over time.

Why MINT is not a substitute for a public model

  • The code is not open.
  • External researchers cannot freely reproduce or extend the model.
  • Access to the underlying data is heavily restricted.
  • The model is designed first for SSA’s analytical mission, not for broad public use.

What this project should learn from MINT

MINT is a reminder that the hardest part of this project is not benefit arithmetic. It is the lifecycle data problem. A public model will not match MINT’s administrative-data advantage in the near term, so it must compensate with unusually explicit validation and unusually clear disclosure of uncertainty.

CBOLT: official fiscal authority, different objective function

CBO’s long-term model, often referred to through the CBOLT framework, serves a different institutional purpose from the one proposed here (Congressional Budget Office 2018, 2004). It should not be caricatured as a representative-agent model. Public CBO materials describe an individual microsimulation model with administrative earnings records, imputed family characteristics, annual transitions, and spouse links used for Social Security benefit calculations (Congressional Budget Office 2018, 2019).

What CBOLT optimizes for

  • internal consistency with CBO’s broader long-term fiscal outlook
  • aggregate budget analysis
  • macro-fiscal coherence across policy domains
  • distributional Social Security results backed by administrative earnings histories

Why that matters

This comparison helps discipline the project’s claims. A public dynamic microsimulation model should not pretend to outcompete CBO on its own institutional objective function. The comparative advantage here is micro-distributional analysis with transparent assumptions, not official budget authority.

PWBM: public-facing outputs without full public model access

The Penn Wharton Budget Model shows that public-facing interfaces can be powerful even when the underlying system remains relatively closed. It is a useful reminder that accessibility has layers:

  • public-facing tools can broaden reach
  • but limited code and data transparency still constrain independent verification

The implication for this project is straightforward: a web interface is helpful, but it is not a substitute for an open modeling pipeline.

The Cato model: the nearest open system

The Cato Institute’s open-source Social Security model simulates mortality, fertility, marriage, divorce, and employment as stochastic transitions on roughly 10,000 households from the 2007 CPS ASEC, and reports trust-fund metrics and reform scores (Chanwong 2026). It focuses on OASI without a separate SSDI module, holds its labor-force transition matrix constant from 2024 onward, and its repository publishes no validation against SSA Trustees, MINT, or DYNASIM projections (see the fuller characterization below and in the funder summary).

Its existence raises the bar. This project cannot claim novelty just because it is open. The differentiators have to be stronger than that:

  • tighter integration with PolicyEngine’s tax-benefit ecosystem
  • a larger and better-calibrated public base population
  • a stronger validation story
  • a clearer path from research pipeline to public interface

International dynamic models

The pattern of closed national models repeats internationally — Pensim2 at the United Kingdom’s Department for Work and Pensions, MOSART at Statistics Norway, MIDAS at Belgium’s Federal Planning Bureau (Li and O’Donoghue 2013) — but the exceptions mark the direction of travel:

  • Destinie 2 (INSEE, France): the pension microsimulation model behind France’s official projection exercises, with source code published under the GPL (Blanchet et al. 2010) (github.com/InseeFr/Destinie-2). A national statistical institute releasing its production dynamic model is the strongest precedent for the access model this project proposes.
  • SimPaths (CeMPA, University of Essex): an open-source life-course microsimulation model estimated for the United Kingdom, with adaptations underway for several other European countries (Bronka et al. 2025). The closest international analogue on openness, and proof that a research-grade open ecosystem can form around dynamic modeling.
  • Open frameworks without open populations: LIAM2, developed at Belgium’s Federal Planning Bureau (Menten et al. 2014), and OpenM++, an open reimplementation of Statistics Canada’s Modgen platform (OpenM++ development team 2026), supply generic dynamic-simulation engines. Neither ships a calibrated population, a maintained rules stack, or a scoring protocol; they solve the engine problem, not the credibility problem. WIFO’s microWELT, built on OpenM++, models welfare transfers comparatively across Austria, Spain, Finland, and the United Kingdom, with a United States variant for labor-force projection (Spielauer et al. 2020) — one open engine carrying a multi-country dynamic model, still without a certified baseline or a scoring program.

What none of these combines is an open codebase, a certified calibrated microdata baseline, and a public scoring protocol under which the model’s claims resolve. That combination — not openness alone — is the gap.

Comparison table

Model Main strength Main limitation What we should learn
DynaSim Rich retirement-policy scope and long validation history Not open or easily accessible Match the rigor, not the access model
MINT Administrative-data credibility Restricted data and code Treat validation as the central challenge
CBOLT Official long-term fiscal authority plus administrative earnings credibility Limited public reproducibility of the production pipeline Do not overclaim on official scoring, administrative-data parity, or macro closure
PWBM Public-facing policy communication Limited transparency beneath the interface Public tools help, but they are not enough
Cato model Open-source; simulates ~10,000 ASEC households under SSA assumptions No published validation against administrative benchmarks Openness alone is not the differentiator
SimPaths Open-source life-course modeling, spreading across European countries No published resolution-scoring program An open research ecosystem is achievable
Destinie 2 Official production model with published GPL source France-specific scope; no published scoring program Governments can open production models

What gap still exists

No current model combines all of the following:

  • open source
  • a public-data workflow
  • transparent intermediate-state validation artifacts
  • a programmatic and AI-callable API
  • a public interface
  • integration with a broader tax-benefit platform

That is the real gap this project fills.

What this project should and should not claim

Plausible claims

  • it can become the most transparent dynamic Social Security model in the U.S. policy ecosystem
  • it can reduce the cost of serious exploratory policy analysis
  • it can create a public validation benchmark for lifecycle microsimulation
  • it can eventually support adjacent domains that share the same longitudinal data bottleneck

Implausible or premature claims

  • that it will quickly replace MINT for SSA’s purposes
  • that it will match CBO’s institutional role
  • that public-data reconstruction eliminates the value of administrative records
  • that an interface alone creates credibility

Bottom line

Existing models already show what a serious Social Security microsimulation model looks like. DynaSim shows the value of breadth and institutional memory. MINT shows the value of administrative earnings histories. CBOLT shows what official fiscal authority optimizes for. Open models show that reproducibility is feasible.

The point of this project is to assemble a different bundle of strengths: public reproducibility, integration with PolicyEngine, and a validation record strong enough to make the model useful even before it is trusted as an official benchmark.