Summarise ED intensity statistics — summary.ED • lossratio

Compute group-wise summary statistics for incremental loss intensity \(g\) from an object of class "ED". This function serves two purposes:

Diagnostics: provides descriptive statistics (mean, median, wt, cv) that help the user assess the stability and consistency of observed \(g\) values across cohorts.
Estimation: fits a no-intercept weighted least squares model per development link to produce the WLS-estimated intensity (g), its standard error (g_se), relative standard error (rse), and residual sigma (sigma). These are used downstream by fit_ed().

Usage

# S3 method for class 'ED'
summary(object, alpha = 1, digits = 5, ...)

Arguments

object: An object of class "ED", typically produced by build_ed().
alpha: Numeric scalar controlling the variance structure in the WLS fit. Default is 1.
digits: Number of decimal places to round numeric columns. Default is 5. Pass NULL to skip rounding.
...: Additional arguments passed to the internal WLS estimation.

Value

A data.table with class "EDSummary" containing one row per development link with descriptive statistics and WLS estimates.

Relationship between `wt` and `g`

Both wt and g are weighted averages of the observed intensities, but they differ in how weights are assigned:

wt: Exposure-weighted mean: \(wt = \sum \Delta C^L_{i,k+1} / \sum C^P_{i,k}\). Computed from all rows where both values are finite. Independent of alpha.
g: WLS-estimated intensity from lm(delta_loss ~ exposure_from + 0). Only rows where exposure_from > 0 are used. When alpha = 2, g and wt are numerically equivalent.

See also

build_ed(), fit_ed()