Skip to contents

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.

Usage

chain_ladder(
  triangle,
  average = c("volume", "simple", "regression"),
  periods = NULL,
  exclude = NULL,
  tail = 1,
  tail_from = 1,
  sigma = c("log-linear", "Mack"),
  tail_periods = 100
)

Arguments

triangle

A numeric matrix of cumulative claims: one row per origin period, oldest first, and one column per development period, with NA for the future. A data frame is converted with as.matrix(); see as_triangle to 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 (TRUE to leave out), or NULL (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 over tail_periods periods.

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

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