1
-
Copy the MacroMod URL
+
Copy the PolicyEngine Macro URL
Click the copy button — you'll paste it in the next step.
https://policyengine--macromod-mcp-serve.modal.run/mcp
2
Open Settings → Connectors
-
In ChatGPT, enable Developer mode under Connectors, then Create a connector named MacroMod and paste the URL.
+
In ChatGPT, enable Developer mode under Connectors, then Create a connector named PolicyEngine Macro and paste the URL.
3
@@ -274,7 +274,7 @@
Using Codex?
-
Nine tools across the four engines: score_reform and list_reform_variables (OBR); forecast_uk, latest_shocks, model_summary (SVAR); calculate_household, household_reform_impact, list_reform_parameters (PolicyEngine); and og_score_reform_steady_state (OG-UK — local CLI only: a solve takes tens of minutes, so the hosted server excludes it). The server is serverless and scales to zero — the first call after a quiet spell takes ~10 seconds to wake.
+
Nine tools across the four engines: score_reform and list_reform_variables (OBR); forecast_uk, latest_shocks, model_summary (SVAR); calculate_household, household_reform_impact, list_reform_parameters (PolicyEngine — household-level); and og_score_reform_steady_state (OG-UK — local CLI only: a solve takes tens of minutes, so the hosted server excludes it). Two things stay local-only and are never on the hosted server: OG-UK scoring, and PolicyEngine population-level reform scoring (it needs large private microdata) — the PolicyEngine tools above are household-level only. The server is serverless and scales to zero — the first call after a quiet spell takes ~10 seconds to wake.
@@ -750,7 +750,9 @@
Sector output, long-run
tax, National Insurance, Universal Credit, SNAP, EITC, and the rest —
for the household you describe. Population-level analysis over
representative microdata is in the same package; population-level
- reform scoring through MacroMod is planned.
+ reform scoring is available locally (it needs large private microdata),
+ but is not on the hosted MCP server — the hosted PolicyEngine tools are
+ household-level only.
@@ -777,7 +779,7 @@
Sector output, long-run
@@ -116,7 +119,8 @@
Four model classes, side by side.
from first principles and answers where the economy settles in the long
run; one reproduces the official forecaster's empirical system and
answers what happens over the next few years; one imposes minimal
- theory and answers what is driving the economy right now; and one
+ theory and answers what is driving the economy at the latest
+ available data vintage (currently 2024 Q2); and one
drops from aggregates to people, computing the actual statute for a
specific household.
@@ -147,10 +151,24 @@
Match the model to the question.
- "How does this tax change affect growth and the deficit over the next 3–5 years?" — the OBR model. It's built for near-term fiscal scoring and speaks the same language as the official forecast.
- "What does this reform do to work incentives, saving, and the long-run size of the economy?" — the OLG model. It captures how households of every age re-optimise over their lifetimes.
- - "What's driving GDP and inflation right now — and why did the forecast change since last quarter?" — the structural VAR. It decomposes the current data into named structural shocks, forecasts with credible bands, and splits a forecast revision into news and reassessment.
+ - "What's driving GDP and inflation at the latest data vintage — and why did the forecast change since the quarter before?" — the structural VAR. It decomposes the data at the latest available vintage (currently 2024 Q2) into named structural shocks, forecasts with credible bands, and splits a forecast revision into news and reassessment.
- "What does this reform mean for a nurse on £35,000 with two kids?" — PolicyEngine. It runs the actual UK or US statute for that exact family and reads off taxes, benefits, and net income, before and after.
- "I want both the transition and the destination." — run both scoring models. They report comparable real-world aggregates, so you can read the near-term multiplier against the long-run equilibrium.
+
+ Verification gradient. The models also differ in how far
+ they can be checked against ground truth, and that should temper how much
+ weight you put on each answer. The OBR emulator and the
+ structural VAR are replications with published
+ anchors — the OBR's Economic and Fiscal Outlook tables and the Bank of
+ England paper's figures, respectively — so their output can be validated
+ against a known benchmark. OG-UK is a
+ structural counterfactual model with no ground truth: it is
+ calibrated to national-accounts targets, not validated against a published
+ result. PolicyEngine computes statute directly, checkable rule by rule
+ against the tax and benefit code. A replication you can audit against its
+ anchor; a calibrated counterfactual you can only reason about.
+
Running the same reform through both scoring models is the point of a suite:
agreement raises confidence, and a divergence is informative — it usually
@@ -203,7 +221,7 @@
Pick a model and score a reform.