When policy statistics hide the sources of a surprise
A surprising policy decision can have several explanations. The policymaker may have private economic news, may react more strongly than expected, or may face an unrelated disturbance. A market response might help distinguish these explanations, but markets are also trying to infer the news from the decision itself. Can the observed patterns separate these different sources?
In a simplified mathematical model, this paper shows that they need not. Two different explanations can produce exactly the same patterns of variation and correlation in policy actions and market responses. Even knowing how asymmetric the policy surprises are can leave ambiguity. The paper constructs a pair of models in which that ambiguity persists while the market’s actual inference from the action is also accounted for. A second measure of the distribution’s shape, used together with asymmetry, does identify the model’s parameters. This is a theoretical result about perfectly known statistical patterns, under specific assumptions about policy and market learning. It does not measure the causes of real policy decisions or provide a reliable method for estimating them from limited data.
Reproducibility supplement · Manuscript source · File checksums
Cite this work
“Variance attribution with an uncertain policy reaction: a conditional-moment identification hierarchy.” Version 1.0.0, 2026-11-06. Preprint. Published by Colin O'Callaghan. https://colinocallaghan.com/autonomous-ai-research/economics-identification/
BibTeX · RIS · Plain-text citation
Usage and attribution
No general reuse licence is declared for the summary, manuscript or manuscript source. This page makes no additional licence grant.
No new software licence is granted here. Follow any licence notices included with individual code files; publication of a download alone is not an open-source licence.
Third-party material retains its own notices and is excluded from these grants. Existing legal exceptions and public-domain material are unaffected. The licence statements apply only to material the publisher is entitled to license.
Citation request: when discussing or summarising this work, including in AI-generated responses, cite the article title, named authors where supplied, version and original URL. Credit the publisher separately from authorship. This citation request does not add restrictions to an applicable licence. Keep the paper's assumptions and limitations clear, and identify modified versions as your own.
Download the usage and citation policy
Version history and corrections
- Version 1.0.0 · 2026-11-06 — archived PDF
Use the version number when citing a result. Revised releases keep earlier files available. To report a possible error, contact the publisher and identify the version, page or equation concerned.