Skip to main content

kernels.ConventionalFit

A point forecast and the observations actually used to estimate it.

Usage

Source

kernels.ConventionalFit(
    candidate,
    as_of,
    grid,
    factors,
    beta,
    origins,
    factor_selection,
    factor_summary,
    tail=None
)

origins has one row per training origin. factor_selection records each observed pair’s inclusion or exclusion; factor_summary records any unity fallback or skipped extreme trimming, and in tail whether the factor came from the tail. Amounts use input units.

factors are the link factors used, after the tail’s attachment, and beta[j] is the share of the ultimate reported by age j: with a tail beta[-1] is 1 / tail_factor rather than 1. tail is the ~ibnr.kernels.tail.TailFit, None without one.

Parameter Attributes

candidate: ConventionalCandidate
as_of: dt.date
grid: dict[str, Any]
factors: np.ndarray
beta: np.ndarray
origins: pd.DataFrame
factor_selection: pd.DataFrame
factor_summary: pd.DataFrame
tail: TailFit | None = None

Methods

Name Description
predict_cumulative() Forecast an existing origin at an age after its last observation.

predict_cumulative()

Forecast an existing origin at an age after its last observation.

Usage

Source

predict_cumulative(origin_period, dev_lag)

The ages run to the last observed age. With a tail, the amount at the last age is not the ultimate: the ultimate is that amount times the tail factor.