working paper · October 2026

Microcosm Dynamics: an open, scored dynamic microsimulation model

Microcosm Dynamics follows individual Americans through time and computes each person's Social Security benefits year by year. From the Panel Study of Income Dynamics, it learns how people earn, marry, divorce, have children, become disabled and die, and it draws claim ages from the Social Security Administration's published awards. Its earnings, family and work-limiting disability components pass tests on held-out or external data, as do its transitions once carried onto the Current Population Survey, and the whole engine, fitted through 2014, passes a 2015–2019 holdout. Mortality and claiming run without such a test, and four of the passes rest on public amendments we adopted after an earlier attempt failed. Four blind tests compare the model with DYNASIM3's published results, and none of them uses the tested earnings, family or work-limiting disability components. Two of the blind tests project PSID families to 2030, and those projections draw no earnings after 2010, apply no retirement earnings test and model no claiming response to reform.

population

The model is built to start from Microcosm's certified, representative CPS frame: 166,302 people in 57,240 households, with a 2024 reference period. No projection starts there yet. Every projection so far starts from PSID members, and all four blind tests against DYNASIM3 draw their people from the PSID.

gates

Earnings, family, work-limiting disability and the transport of PSID-estimated processes onto the CPS frame each pass a gate: a test on held-out or external data that we register publicly before we run it, with its pass rule fixed in advance and changed only by public amendment. Mortality, claiming, disability-insurance entitlement and the immigrants who enter the population on a schedule run without one.

data limits

In 2023 the PSID interviewed 9,152 families, against about 95,000 households a year in the Current Population Survey. It follows the families of 1968 and their descendants. About 40 percent of the surviving 1968 sample still responded in 2015. The PSID's records of low earnings have known measurement problems: in a study of men's earnings, about 5 percent of PSID observations had earnings under $20,000 a year, against about 25 percent in the Census Bureau's Longitudinal Employer-Household Dynamics records. Section 6 of the paper sets out what the model does about each limit, including a check against the Social Security Administration's public file of earnings records, which so far has scored the model's rules for filling gaps in recorded careers but not its generated earnings or whole careers.

blind tests

A one-point cut to cost-of-living adjustments, which acts mechanically on benefits already being paid, lands within 0.74 percentage points of DYNASIM3 in every age group, so the match says little about how well the model captures how lives change. Each of the other three tests finds a gap between the two models.

next

The forward earnings law failed B2, the registered test that carries it from 2010 to 2014, in October 2026, so a successor law, designed on evidence other than that result, comes first. We then plan to connect forward earnings to the projection tests, apply the retirement earnings test, add behavioral claiming and run the full benefit rules in the Axiom rules engine. So far Axiom reproduces one piece of the model's Python benefit rules: average indexed monthly earnings computed over the number of years the law sets, while the projection tests count a fixed 35 years.