Simulates the annual losses of several lines of business, each with its own frequency and severity, and joins them with a Gaussian or t copula, for the capital of the portfolio: its value at risk and tail value at risk, the diversification between the lines, and each line's share of the capital.
Usage
simulate_portfolio(
lines,
correlation = 0,
copula = c("gaussian", "t"),
df = 4,
n = 10000,
level = 0.995,
seed = NULL
)Arguments
- lines
A named list with one element per line: a list of arguments of
simulate_claims(withoutn_simsandseed), whosetotal_claimsare the line's losses, or a numeric vector ofnsimulated losses.- correlation
The correlation between the lines' copula draws: a matrix with a row and a column per line, symmetric and positive definite with ones on the diagonal, or one number for every pair of lines.
- copula
"gaussian"(the default) or"t".- df
The degrees of freedom of the t copula: the lower, the stronger the dependence in the tail.
- n
The number of simulations.
- level
The level of the value at risk and the tail value at risk, such as 0.995 (the default) for 1 in 200 years.
- seed
A whole number for a reproducible run, or
NULL(the default) to use the current random number stream. The stream of the R session is restored after a run with a seed.
Value
An object of class netsimr_portfolio, a list with losses, a
data frame with a column per line and the total, one row per simulation;
summary, a data frame with the mean, standard deviation, VaR and TVaR of
each line and of the total, each line's capital (co-TVaR) and its share;
diversification, the benefit in amount and as a share of the lines' VaR;
and the copula settings.
Details
Each line is simulated on its own, by simulate_claims (or
given as a vector of simulated losses). The lines are then joined by
reordering: n draws from the copula give, for each line, the rank its
losses take in each simulation, and the line's simulated losses are put in that
order. Each line keeps exactly the distribution it was simulated with, and the
ranks of the lines follow the copula: with the Gaussian copula, normal draws
with the given correlation matrix; with the t copula, the same divided by a
shared \(\sqrt{W / \nu}\), \(W\) chi-squared with \(\nu\) degrees of
freedom, which makes large losses in several lines at once more likely (tail
dependence). This is the reordering used in capital models (the method of
Iman and Conover with the copula's ranks).
The value at risk of the total at level \(p\) is its \(p\) quantile, and the tail value at risk the mean of the totals at or above it. The diversification benefit is the sum of the lines' VaR less the total's VaR. Each line's capital is its mean loss in the simulations whose total is at or above the total's VaR (co-TVaR, the Euler allocation of the TVaR), so the lines' capitals add up to the total's TVaR.
References
Iman, R. L. and Conover, W. J. (1982). A distribution-free approach to inducing rank correlation among input variables. Communications in Statistics - Simulation and Computation 11(3), 311-334.
Examples
lines <- list(
motor = list(frequency = "Poisson", frequency_params = 50,
severity = "LogNormal", severity_params = c(8, 1)),
property = list(frequency = "Poisson", frequency_params = 5,
severity = "Pareto", severity_params = c(1.8, 20000))
)
portfolio <- simulate_portfolio(lines, correlation = 0.3, n = 20000, seed = 1)
portfolio
#> Portfolio of 2 lines, 20,000 simulations, Gaussian copula
#>
#> line mean sd VaR 99.5% TVaR 99.5% capital share
#> motor 245,673 57,400 431,722 474,597 307,704 0.134
#> property 223,855 217,936 1,122,307 1,986,134 1,982,527 0.866
#> total 469,528 236,837 1,410,849 2,290,232 2,290,232 1.000
#>
#> Diversification benefit: 143,180 (9.2% of the lines' VaR of 1,554,029)
# stronger dependence in the tail with a t copula
tail <- simulate_portfolio(lines, correlation = 0.3, copula = "t", df = 3, n = 20000, seed = 1)
tail$diversification
#> stand_alone_var total_var benefit benefit_share
#> 1.554029e+06 1.444839e+06 1.091902e+05 7.026266e-02
