US recession-fragility simulation — decision-grade macro stress test
A self-built economic simulator (Dyno-Sim) that stress-tests how fragile the US is to a 2026 recession from the 2025 starting point — and answers a real decision question (does the fiscal-monetary policy stack hold the system out of crisis, and with what confidence) with every line traceable to a run, an equation, and a validation test.
Outcome metrics
- Recommendation-flip (how often a plausible shock flips the stance)
- 47.0% [46.3–47.7%] (Wilson CI, 20k replicas)
- Calibrated GFC fit to real FRED data (employment + policy rate)
- combined RMSE 0.0876 → 0.0372 after calibration
- Validation profiles passing the rule-gate
- 38 / 38
- Decision artifact
- Simulation Decision Pack — traceable to run + equation + test
The question
"How fragile is the US to a 2026 recession given the 2025 starting point (≈4.3% policy rate, private debt ≈1.5× GDP), and does the policy stack hold the system out of crisis — with what confidence?" That is a decision question, not a chart request. Comuvia answers it with a simulation, not an opinion.
What it produces
- A central case. Off the elevated-rate, high-private-leverage 2025 base, a moderate recession tips the private sector into a debt-deflation dynamic: employment stays depressed and private debt climbs to ~2.1× GDP. The policy stack stabilises the public balance sheet (rate cut + QE + fiscal support) but cannot arrest the private-debt spiral. The fragility lives in the private balance sheet — that is the decision-relevant finding.
- An uncertainty layer. From a 20,000-replica, real-data-calibrated shock surface, the recommendation-flip is ~47%: under plausible shock variation, nearly half of replicas would flip the recommended stance. The central case is not robust — and the deliverable says so, in a number.
- Calibration to real history. The model is fit to real FRED EFFR + UNRATE over the GFC and COVID windows. A documented calibration closed a Great-Depression-scale overshoot in the GFC fit (employment trough 0.72 → 0.91, matching the real ~10% unemployment) while keeping the policy rate hitting the 2009 zero lower bound — combined error cut more than half.
What it proves
This is Comuvia's verification-centric method made concrete: the model proposes; an owned verifier decides. Every claim is checkable against ground truth — equations, real-data validation profiles, and a rule-gate that all pass. The exit artifact is a Decision Pack where each line traces to a run id, an equation, and a test, so a decision-maker can audit why, not just read a recommendation.
Honest constraints
A single short rate (no term structure); canonical empirical policy rules, not a fitted DSGE; one illustrative scenario, not a calibrated point forecast; the recommendation-flip is the honest uncertainty around it. Research / paper-only. Not investment advice. Not a forecast.
Why it matters for a buyer
For an allocator, family office, or PE owner, the value is decision-grade and auditable: a transparent, vendor-neutral model that states what it does not know (the 47% flip) as clearly as what it does — the opposite of a black-box score.