kernels.tail_variance()
The tail step’s sigma and the tail factor’s standard error, for Mack.
Usage
kernels.tail_variance(
f,
sigma2,
s,
tail_factor,
*,
sigma=None,
std_err=None,
dev_grain_months=12,
option_prefix=""
)Mack (1999) and R’s MackChainLadder(tail=): the tail is one more development step with factor tail_factor. Its position p on the link axis is where a straight line through (t, log(f_t - 1)), over the factors above 1, reaches log(tail_factor - 1). The tail’s sigma is read at p off a straight line through (t, log sigma_t) over the positive sigmas, and its standard error off one through (t, log se_t), with se_t = sqrt(sigma2_t / s_t), over the positive ones.
These are R’s lines, with the points that have no logarithm left out. chainladder-python leaves such a point’s t in its sums (and fills a zero sigma with 1e-320), so it agrees only where no factor is at or below 1 and no sigma is zero or missing.
sigma and std_err replace their own read; the position is found only when one of them is still needed. A tail factor of exactly 1 gives zero for both reads. Refused as variance_not_estimable: a tail factor below 1 without both values given, fewer than two factors above 1 or two positive sigmas (or standard errors) when a line is needed, factors that do not decay (a line that does not fall), and a position below 1, before the first link, where the sigma would be extrapolated backwards.