Skip to contents

Fit stage-adaptive (SA) loss projection on a Triangle

Source: R/sa.R
fit_sa.Rd

Project cumulative loss across the cohort x development grid using the stage-adaptive (SA) method: ED before the maturity point, CL after. SA composes both projection paradigms anchored on a per-group maturity switch – a 2-pass fit (maturity detection via fit_ata(), then the SA projection itself).

SA is a worker – standalone, no internal method dispatch. The role-specific entry point fit_loss() dispatches method = "sa" to this function; users can also call fit_sa() directly.

Usage

fit_sa(
  x,
  loss = "loss",
  exposure = "premium",
  alpha = 1,
  exposure_fit = NULL,
  exposure_method = c("ed", "cl"),
  exposure_alpha = 1,
  sigma_method = c("locf", "min_last2", "loglinear", "mack", "none"),
  recent = NULL,
  regime = NULL,
  maturity = "auto",
  tail = FALSE,
  conf_level = 0.95,
  bootstrap = NULL,
  B = 999L,
  seed = NULL,
  type = c("parametric", "nonparametric", "analytical")
)

Arguments

x

A "Triangle" object. The standardized "loss" and "premium" columns are used (as_triangle() produces these).

loss

Cumulative loss column name. Default "loss".

exposure

Cumulative premium column name. Default "premium".

alpha

Variance-structure exponent for the loss fit. Default 1.

exposure_fit

Optional pre-built PremiumFit supplying the premium projection. When NULL, fit_sa() builds the premium projection internally – a worker-layer fit_cl() on the premium column – using exposure_method, exposure_alpha, and the resolved regime.

exposure_method

One of "ed" (default) or "cl". Used only when exposure_fit = NULL.

exposure_alpha

Variance-structure exponent for the premium fit. Default 1.

sigma_method

Method used to extrapolate sigma for links where it cannot be estimated. One of "locf" (default), "min_last2", "loglinear", "mack", or "none". "mack" applies the Mack (1993, Appendix B) tail estimator to the last unestimated link only, falling back to LOCF for any earlier ones with a warning. "none" performs no extrapolation; sigma stays NA and downstream variance terms drop those links via finite-value guards. Passed to .extrapolate_sigma_ata().

recent

Optional positive integer; calendar-diagonal filter.

regime

Optional regime specification (loss-side). Accepts the standard 4-type dispatch (NULL / Regime / "auto" / function). In SA mode the resolved regime drives the hybrid 2-pass filter (cohort cut for the ED phase, calendar-diagonal wedge for the CL phase).

maturity

Maturity specification. Default "auto". Accepts the standard 4-type dispatch (NULL / Maturity / "auto" / function). SA requires a maturity – NULL disables SA entirely (use ED or CL directly in that case).

tail

Logical or numeric; tail factor for the CL phase. Forwarded to the internal premium fit when relevant.

conf_level

Confidence level for the analytical CI on the loss projection. Default 0.95.

bootstrap

Bootstrap configuration (NULL / TRUE / FALSE / "auto" / BootstrapTriangle / lazy function). Default NULL resolves to "auto" (residual bootstrap) for SA.

B

Integer number of bootstrap replicates. Default 999.

seed

Optional integer seed.

type

Bootstrap process type. Default "parametric". (Only used when bootstrap = "auto".)

Value

An object of class "SAFit". List with components mirroring LossFit: full, proj, maturity, loss_ata_fit, premium_ata_fit, premium_fit, ed, factor, selected, plus metadata (method = "sa", alpha, sigma_method, recent, regime, conf_level, ci_type, bootstrap, usage).

See also

fit_loss(), fit_cl(), fit_ed(), fit_ratio().

Examples

if (FALSE) { # \dontrun{
data(experience)
tri <- as_triangle(
  experience[coverage == "surgery"],
  groups   = "coverage",
  cohort   = "uy_m",
  calendar = "cy_m",
  loss     = "incr_loss",
  premium = "incr_premium"
)

sa <- fit_sa(tri)
summary(sa)
} # }