consensus_bias

consensus_bias(
    estimates,
    as_of,
    *,
    process_noise=defaults.CONSENSUS_BIAS_PROCESS_NOISE,
    measurement_noise=defaults.CONSENSUS_BIAS_MEASUREMENT_NOISE,
    source=CONSENSUS_SOURCE,
)

Estimate each entity’s persistent consensus error from its own history.

A period’s relative error compares the consensus that stood strictly before that period’s actual became knowable with the actual. A local-level filter runs over those errors oldest first, and each subject reads the filtered state after the latest realised period strictly before its own period_end.

Parameters

Name	Type	Description	Default
estimates	pd.DataFrame	Long input panel.	required
as_of	pd.Timestamp	The moment to read the panel at.	required
process_noise	float	Nonnegative process variance of the filter.	defaults.CONSENSUS_BIAS_PROCESS_NOISE
measurement_noise	float	Positive measurement variance of the filter.	defaults.CONSENSUS_BIAS_MEASUREMENT_NOISE
source	str	The source whose bias is estimated.	CONSENSUS_SOURCE

Returns

Name	Type	Description
	pd.DataFrame	One row per subject present in the panel as of as_of, carrying consensus_bias, as a fraction of the actual with positive meaning consensus too high, and consensus_bias_count, the errors behind it. The bias is NaN before the first error; the count is always the true one. A filter variance out of range raises PanelError.