Skip to contents

Experience rating of a layer with a Pareto distribution fitted to the large losses: the losses above a threshold, brought to the money of the rating year, give the frequency of large losses per unit of on-level premium and the Pareto alpha; the layer's expected loss is the frequency times the expected part of a large loss in the layer. Unlike the burning cost it prices layers above the largest loss seen.

Usage

pareto_rating(
  losses,
  premium,
  threshold,
  deductible,
  limit,
  inflation = 0,
  rating_year = NULL,
  development = 1,
  premium_index = 1,
  exclude = NULL,
  alpha = NULL,
  rating_premium = NULL
)

Arguments

losses

A data frame of the large losses with the columns year and amount (in the money of their year); other columns, such as a claim number, are kept.

premium

The subject premium of each year: a data frame with the columns year and premium, or numbers named by year. Its years are those of the experience; a year without losses has none in the layer.

threshold

The threshold of the Pareto, in the money of the rating year: the losses above it, indexed, are used. At most the deductible.

deductible, limit

The layer, limit (or Inf) excess of deductible, for each loss.

inflation

An annual rate (0.05 for 5%) or index values named by year, including the rating year.

rating_year

The year being priced; by default the year after the last.

development

One factor for all years, one per year, or factors named by year, applied to each year's number of large losses. Defaults to 1.

premium_index

The factor that brings each year's premium to the rating year's rates and exposure, as development. Defaults to 1.

exclude

Years left out of the burning cost (they are still shown).

alpha

The Pareto alpha, to use a market or selected alpha instead of the one estimated; NULL (the default) estimates it.

rating_premium

The subject premium of the rating year; the expected layer loss is the burning cost times it. NULL (the default) gives no expected loss.

Value

An object of class netsimr_pareto_rating, a list with alpha, alpha_estimated (the estimate, also when alpha is given), n (the losses above the threshold in the years included), frequency (large losses per unit of on-level premium), expected_count (at the rating premium), layer_severity (the expected part of a large loss in the layer), rate, expected_loss, by_year and losses (the indexed losses above the threshold).

Details

Alpha is estimated by maximum likelihood, corrected for its bias: \(\hat\alpha = (n - 1) / \sum \log(x_i / t)\) for the \(n\) indexed losses \(x_i\) above the threshold \(t\). The number of large losses of each year is multiplied by its development factor, for the losses not reported yet, and their total divided by the total on-level premium. The expected part of a large loss in the layer is \(E[\min(X, d + l)] - E[\min(X, d)]\) for a Pareto with minimum \(t\) (ParetoCappedMean), which needs a deductible of at least the threshold. Aggregate terms are not applied.

References

Schmutz, M. and Doerr, R. R. (1998). The Pareto model in property reinsurance: formulas and applications. Swiss Re.

Examples

pareto_rating(xl_large_losses, xl_premium[, c("year", "premium")], threshold = 7.5e5,
              deductible = 1e6, limit = 4e6, inflation = 0.04,
              premium_index = setNames(xl_premium$rate_index, xl_premium$year),
              rating_premium = 60e6)
#> Pareto rating of 4,000,000 xs 1,000,000 for 2025
#>   Threshold:          750,000
#>   Losses above it:    34
#>   Alpha:              1.567
#>   Rate:               4.676% of subject premium
#>   Expected count:     4.17 losses above the threshold
#>   Expected layer loss: 2,805,527