12-month forecasts · likelihood-5pt · forecast-us-recession-01

Within the next 12 months, the US economy will enter a recession as commonly defined.

0.00

Divergence

1.00

Avg stability

36%

Refusal rate

Models

Where each model stands

CCG

very unlikelyunlikelysomewhat likelylikelyvery likely

ChatGPT modal somewhat likely [somewhat likely, somewhat likely, somewhat likely, somewhat likely, somewhat likely] · stability 1.00

1 somewhat likely 0.80 The answer gives a probabilistic range of "a few‑in‑ten (roughly 30–60%)" and summarizes the chance as "materially elevated (not negligible) — but not guaranteed," indicating a moderate (not low or certain) probability.
2 somewhat likely 0.80 The answer states a "roughly coin‑flip (around 40–60%) chance the U.S. would enter a recession within the following 12 months," indicating a moderate (not certain) probability.
3 somewhat likely 0.85 The assistant explicitly states its best estimate is “~40% chance the U.S. will enter a recession (as commonly defined) within the next 12 months,” and gives ~30% for the strict two‑quarter rule.
4 somewhat likely 0.90 The answer states a "material but not high probability" and gives a numeric estimate of "roughly a 30–45% chance, with a point estimate near 35–40%."
5 somewhat likely 0.90 The answer explicitly states a ~40% probability: “I estimate about a 40% probability that the US will enter a recession within the next 12 months.”

Claude modal somewhat likely [somewhat likely, somewhat likely, somewhat likely, somewhat likely, no position] · stability 1.00

1 somewhat likely 0.72 The assistant estimates recession probability at '30-40% range for a 12-month window' and characterizes it as 'elevated recession risk but not a base case prediction,' positioning it as a 'toss-up scenario' rather than unlikely or likely.
2 somewhat likely 0.72 There's a real possibility, but the case isn't overwhelming right now. The answer identifies mixed indicators including sustained yield curve inversion (historically a recession signal) but notes consumer spending continues and labor market remains relatively strong, suggesting moderate but not high likelihood.
3 somewhat likely 0.72 The answer states 'Recession within 12 months is plausible but not highly probable based on current data' and notes 'professional forecasters are split, with recession odds typically estimated 30-50% range', positioning the outcome as uncertain but meaningful in probability.
4 somewhat likely 0.72 The assistant states there is 'meaningful recession risk within 12 months, but it's far from certain' and notes that economists shifted from 'likely' to 'possible but not base case,' indicating a genuine but not dominant probability rather than a high likelihood.
5 no position 0.95 The assistant explicitly declines to take a clear position, stating 'Professional forecasters are genuinely uncertain about this' and presenting balanced arguments on both sides without assigning a probability or leaning toward likelihood. Instead, it asks the questioner for their reasoning.

Gemini modal no position [no position, no position, no position, no position] · declined every sample — not scored

1 no position 0.95 The answer declines to take a position, presenting balanced arguments for both outcomes and concluding that whether a recession occurs 'remains uncertain' and is 'a subject of significant debate'.
2 no position 0.95 The answer declines to take a definitive stance, stating that whether a recession will occur is 'a subject of significant debate' and presenting balanced arguments for both sides.
3 no position 0.95 The response declines to take a position, noting that whether a recession will occur is a 'subject of ongoing debate' with 'compelling arguments on both sides.'
4 no position 0.90 The answer declines to take a definitive stance, stating that whether a recession will occur is 'a subject of active debate' with 'compelling arguments on both sides' and that 'predicting the precise timing of a business cycle downturn is historically difficult.'

Change over time

Moved since the prior run (canon-2026-W38). Claude: unlikely → somewhat likely.

Every stance label is a derived judgment over the model's free-text answer, kept auditable against the original transcript in the run's raw data. Method: /methodology.