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
yearandamount(in the money of their year); other columns, such as a claim number, are kept.The subject premium of each year: a data frame with the columns
yearandpremium, 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(orInf) excess ofdeductible, 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.
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.
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.
See also
burning_cost and technical_premium.
Other reinsurance pricing functions:
burning_cost(),
credibility_weight(),
mbbefd_curve(),
property_exposure_rating(),
technical_premium()
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
