Skip to main content

kernels.TweedieSpec

What to fit: the distribution, the link, the design and the projection.

Usage

Source

kernels.TweedieSpec(
    power=1.0,
    link="log",
    origin="factor",
    calendar="none",
    projection="pattern",
    max_iter=100
)

Attributes

power: float

The Tweedie power: 0, or 1 and above (see the module docstring).

link: str

"log" (the default) or "identity".

origin: str

"factor" (the default): one level per origin. "none": no origin term, so every origin shares one set of expected increments.

calendar: str

"none" (the default) or "trend": a linear trend in the calendar period. Only with origin="none": with origin and development factors the trend is a sum of their columns and cannot be estimated. Under the log link with development factors a calendar trend fits the same means as a linear trend across origins, so do not read it as an inflation rate.

projection: str

How an origin’s ultimate is made. "pattern" (the default) is the latest cumulative times the fitted pattern’s development factor from the latest age to the last, as chainladder-python reports it. "increments" is the latest cumulative plus the fitted future increments, as R’s glmReserve reports it. The two agree at power 1 with a log link and origin factors. "increments" needs origin factors: without them every origin would get the same future amounts whatever its size.

max_iter: int
The most Fisher-scoring steps to take (100 by default).