Skip to contents

Build a development structure from experience data

Source: R/triangle.R
build_triangle.Rd

Aggregate experience data into a development structure by grouping, period, and development-period variables. The result contains:

  • cumulative loss and cumulative risk premium,

  • period and cumulative proportions,

  • margin and cumulative margin,

  • profit indicators,

  • loss ratio (lr = loss / rp) and cumulative loss ratio (clr = closs / crp).

The loss ratio is defined as: $$lr = loss / rp$$

where rp represents risk premium (expected loss), not written premium.

Proportion variables are computed within each cohort_var + dev_var cell:

  • loss_prop = loss / sum(loss)

  • rp_prop = rp / sum(rp)

  • closs_prop = closs / sum(closs)

  • crp_prop = crp / sum(crp)

Therefore, for a fixed (cohort_var, dev_var) cell, the proportions sum to 1 across groups. These are useful for examining the composition of each development cell across products or other grouping variables.

Usage

build_triangle(
  df,
  group_var,
  cohort_var = "uym",
  dev_var = "elap_m",
  value_var = c("loss", "rp"),
  fill_gaps = FALSE
)

Arguments

df

A data.frame containing experience data with loss and risk premium.

group_var

Column(s) used for grouping (e.g., product, gender).

cohort_var

Column(s) defining the exposure period (e.g., underwriting year-month, quarter, half-year, or year such as uym, uyq, uyh, uy).

dev_var

Column(s) defining development periods (e.g., months since issue such as elap_m).

value_var

Character vector specifying the value columns, typically c("loss", "rp") where:

  • loss = actual claims (observed loss),

  • rp = risk premium (expected loss).

fill_gaps

Logical; if TRUE, zero-fill missing (group_var, cohort_var, dev_var) cells so that every cohort has a consecutive dev_var sequence. Default FALSE, which raises an error when gaps are detected. Use validate_triangle() to inspect gaps before deciding.

Value

A data.frame with class "Triangle", containing the following derived columns:

n_obs

Number of distinct periods observed

closs, crp

Cumulative loss and cumulative risk premium

loss_prop, rp_prop

Period proportions within each (period, dev) cell

closs_prop, crp_prop

Cumulative proportions within each (period, dev) cell

margin, cmargin

Period and cumulative margin (rp - loss)

profit, cprofit

Profit indicator (factor "pos" / "neg")

lr

Loss ratio (loss / rp)

clr

Cumulative loss ratio (closs / crp)

The returned object also has an attribute "longer" containing a melted long-format version (class = "TriangleLonger").

Examples

if (FALSE) { # \dontrun{
df <- data.frame(
  pd_cd    = rep(c("P001", "P002"), each = 6),
  pd_nm    = rep(c("cancer", "health"), each = 6),
  uym      = rep(as.Date(c("2023-01-01", "2023-02-01", "2023-03-01")), 4),
  elap_m     = rep(1:2, 6),
  loss     = runif(12, 80, 120),
  rp       = runif(12, 90, 110),
  n_policy = sample(50:100, 12, replace = TRUE)
)

res <- build_triangle(
  df,
  group_var   = pd_cd,
  value_var   = c("loss", "rp", "n_policy"),
  cohort_var = "uym",
  dev_var = "elap_m"
)

head(res)
attr(res, "longer")
} # }