statsmodels.tsa.statespace.news.NewsResults.revision_details_by_update

property NewsResults.revision_details_by_update

Details of forecast revisions from revisions, organized by updates

Returns:
detailspd.DataFrame

Index is as MultiIndex consisting of:

  • revision date: the date of the data revision, that results in revision that impacts the forecast of variables of interest

  • revised variable: the variable being revised, that results in news that impacts the forecast of variables of interest

  • observed (prev): the previous value of the observation, as it was given in the previous dataset

  • revised: the value of the revised entry, as it is observed in the new dataset

  • impact date: the date of the impact on the variable of interest

  • impacted variable: the variable that is being impacted

The columns are:

  • revision: the revision (this is revised - observed (prev))

  • weight: the weight describing how the revision affects the forecast of the variable of interest

  • impact: the impact of the revision on the forecast of the variable of interest

Notes

This table decomposes updated forecasts of variables of interest from the revision associated with each revised datapoint from the new data release.

This table does not summarize the impacts or show the effect of revisions. That information can be found in the impacts table.

This form of the details table is organized so that the revision dates / variables are first in the index, and in this table the index also contains the previously observed and revised values. This is convenient for displaying the entire table of detailed revisions because it allows sparsifying duplicate entries.

However, since it includes previous observations and revisions in the index of the table, it is not convenient for subsetting by the variable of interest. Instead, the revision_details_by_impact property is organized to make slicing by impacted variables / dates easy. This allows, for example, viewing the details of data revisions on a particular variable or date of interest.