Projects a triangle of cumulative claims to ultimate with the chain ladder, and estimates the standard error of each origin's reserve and of the total with Mack's distribution-free model, including the tail.
Arguments
- triangle
A numeric matrix of cumulative claims: one row per origin period, oldest first, and one column per development period, with
NAfor the future. A data frame is converted withas.matrix(); seeas_triangleto build one from a table of claims.- average
"volume"(the default) for volume-weighted development factors,"simple"for simple averages of the link ratios, or"regression"for least squares through the origin.- periods
The number of latest calendar periods (diagonals) whose link ratios are used, or
NULL(the default) for all of them.- exclude
Link ratios to leave out: a logical matrix with a row per origin and a column per development factor (
TRUEto leave out), orNULL(the default) for none.- tail
The tail factor from the last development period to ultimate: a positive number (1, the default, for none), or
"exponential"or"inverse power"for a curve fitted to the development factors above 1 and extrapolated overtail_periodsperiods.- tail_from
The first development factor the tail curve is fitted to.
- sigma
How \(\sigma\) is found for a period with a single link ratio:
"log-linear"or"Mack".- tail_periods
The number of periods beyond the triangle over which a tail curve is extrapolated (100, the default). An inverse power curve falls slowly, so its tail grows with this; an exponential one has nearly all of its tail in the first few periods.
Value
An object of class netsimr_chain_ladder, a list with:
- by_origin
a data frame with, for each origin, the latest claims, the share developed (latest over ultimate), the ultimate, the IBNR (ultimate less latest: for a paid triangle, the IBNR and case reserves), its Mack standard error and coefficient of variation, and the process and parameter parts of the standard error;
- total
the same for all origins together;
- factors, cdf, tail
the development factors, the cumulative development factor from each period to ultimate (with the tail) and the tail factor;
- sigma, factor_se
Mack's \(\sigma\) and the standard error of each factor, the tail's last;
- link_ratios, used
the link ratios and which of them the factors use;
- triangle, full_triangle
the triangle and the triangle completed with the projections, with an ultimate column;
- notes
messages on the choices made, such as an extrapolated \(\sigma\).
Details
The development factor from period \(j\) to \(j + 1\) is the weighted
average of the link ratios \(F_{ij} = C_{i,j+1} / C_{ij}\) that are used,
with weights \(C_{ij}^\alpha\): \(\alpha = 1\) for the volume-weighted
average (the usual chain ladder), 0 for the simple average and 2 for the
regression through the origin. Link ratios can be left out with
exclude, or by using only the latest periods calendar periods.
Link ratios from claims that are zero or negative are always left out.
The standard errors follow Mack (1993, 1999): the variance parameter of each
period is \(\sigma_j^2 = \sum w_{ij} C_{ij}^\alpha (F_{ij} - f_j)^2 / (n_j - 1)\),
and the reserve's mean squared error is the sum of the process and the
parameter error, found recursively; the total adds the correlation of the
origins through the shared factors. Where a period has a single link ratio, its
\(\sigma\) is extrapolated: by a log-linear regression on the others
("log-linear", the default, falling back to Mack's rule when that
regression is not significant at 5%), or by Mack's rule
\(\sigma_{n-1}^2 = \min(\sigma_{n-2}^4 / \sigma_{n-3}^2, \sigma_{n-3}^2, \sigma_{n-2}^2)\)
("Mack"). With a tail factor, the tail's standard error and
\(\sigma\) are extrapolated by log-linear regressions to where a curve
through the factors reaches the tail factor. These choices are those of the
ChainLadder package's MackChainLadder(), whose results the function
reproduces.
References
Mack, T. (1993). Distribution-free calculation of the standard error of chain ladder reserve estimates. ASTIN Bulletin 23(2), 213-225.
Mack, T. (1999). The standard error of chain ladder reserve estimates: recursive calculation and inclusion of a tail factor. ASTIN Bulletin 29(2), 361-366.
See also
bornhuetter_ferguson, cape_cod and
bootstrap_chain_ladder; run_shiny_reserving_tool, an
application for all of them.
Examples
cl <- chain_ladder(raa_triangle)
cl
#> Chain ladder
#>
#> origin latest developed ultimate ibnr se cv process_se parameter_se
#> 1981 18,834 1.000 18,834 0 0 0 0
#> 1982 16,704 0.991 16,858 154 143 0.928 104 98
#> 1983 23,466 0.974 24,083 617 592 0.959 451 383
#> 1984 27,067 0.943 28,703 1,636 713 0.436 530 476
#> 1985 26,180 0.905 28,927 2,747 1,452 0.529 1,219 790
#> 1986 15,852 0.813 19,501 3,649 1,995 0.547 1,820 817
#> 1987 12,314 0.694 17,749 5,435 2,204 0.405 2,039 837
#> 1988 13,112 0.546 24,019 10,907 5,354 0.491 4,946 2,051
#> 1989 5,395 0.336 16,045 10,650 6,332 0.595 6,034 1,918
#> 1990 2,063 0.112 18,402 16,339 24,566 1.503 23,464 7,275
#> Total 160,987 0.755 213,122 52,135 26,881 0.516 24,917 10,085
cl$factors
#> 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9
#> 2.999359 1.623523 1.270888 1.171675 1.113385 1.041935 1.033264 1.016936
#> 9-10
#> 1.009217
# simple averages of the latest five years, with a fitted tail
chain_ladder(raa_triangle, average = "simple", periods = 5, tail = "exponential")$total
#> latest developed ultimate ibnr se cv
#> 1.609870e+05 6.745770e-01 2.386488e+05 7.766182e+04 2.765635e+04 3.561126e-01
#> process_se parameter_se
#> 2.408836e+04 1.358767e+04
