Skill: PolicyEngine — policy-conditional distributions

Skill: PolicyEngine — policy-conditional distributions

PolicyEngine is the microsimulation instrument: when a forecast turns on a tax-benefit parameter or a reform's aggregate impact, call it instead of estimating by analogy. It is an explicit MODEL INPUT, never ground truth — the trace says which policy ids ran and treats the output as one evidence stream with its own error bars.

Endpoints

• Household calc + policy metadata: https://api.policyengine.org/us/... (policy by id: /us/policy/<id>; current-law verification = fetch the policy and confirm parameters).
• Economy-wide impacts: /us/economy/<reform_id>/over/<baseline_id>?region=us&time_period=<year> — returns queued→computing→ok; budgetary impact in the result. Queued or errored runs WIDEN the interval; say so rather than waiting silently (the live forecast-api treats this the same way).
• UK mirror under /uk/.

Calibration of PolicyEngine outputs

Static microsim impacts differ from official scores (CBO/JCT) by behavioral and timing effects. Keep a stored ratio/additive prior from past PolicyEngine-vs-official comparisons and apply it, with the adjustment shown in a math step. Cells whose resolution source IS an official score must forecast the official score, not the raw microsim number.

When to call it

• conditional_on references a tax/benefit parameter: simulate both arms.
• Forecasting program aggregates (CTC outlays, SNAP benefits): baseline run
  • trend adjustment.
• Never for series PolicyEngine doesn't model (CPI, claims) — the trace should not contain decorative microsim calls.