mc: Monte Carlo loss simulation

In barryrowlingson/opVaR: Computes operational risk

Description

Performs Monte Carlo loss simulation for a given number of given periods and some fitted distributions for loss frequency and denstiy.

Usage

mc(x, rfun = NULL, period, iterate, nmb = 1000, begin = NULL, end = NULL,
   wknd = TRUE, crt = 0, type = NULL, param = NULL, zero = F, distname = NULL, fit=T,
   flist = c("beta", "cauchy", "chi-squared", "exponential","f", "gamma", "geometric", 
   "log-normal", "logistic", "normal", "weibull", "inverse gaussian"), 

   p = c(0.95, 0.99, 0.999), ...)

Arguments

x	list with two columnes, first with dates of events and second with loss amount
rfun	random generation for the chosen distribution; NULL default
period	could be "days", "weeks", "months" or "quarters" (loss are aggregated by days, weeks, months or quarters respectively)
iterate	could be "years", "quarters", "months" or "weeks"
nmb	number of period iterations
begin	period begin date; if not given, it would be minimum from loss dates
end	period end date; if not given, it would be maximum from loss dates
wknd	whether to have 252 days a year (FALSE) or not (TRUE, then 365), designed only for period = "days"
crt	correction; argument passed to root.period
type	could be "poisson", "binomial", "nbinomial", if not given, it would be chosen (fitted) automatically
param	start parameters (only for rfun not being character, see examples)
zero	should zero losses be drawn (TRUE) or not (FALSE)?
distname	distribution name; if rfun= rbeta, it should be distname = "beta" (it is designed for that case)
fit	logical: fit rfun parameters?
flist	list of distributions "beta", "cauchy", "chi-squared", "exponential", "gamma", "geometric", "log-normal", "logistic", "negative binomial", "normal", "poisson", "weibull", "inverse gaussian"

p	confidence level (default c(0.95, 0.99, 0.999))

...	arguments passed to root.period

Value

table$losses	generated losses
table$q	losses p-quantile
ad	absolute differences between empirical and fitted density values

Note

"log-normal" and "lognormal" are not the same anymore; only "log-normal" would work

See Also

root.period, loss.fit.dist, fitdistr

Examples

data(loss.data.object)


x<- read.loss(3,2,loss.data.object)

# first example:

 l1 = mc(x,begin="2010-01-01",end="2010-12-31")$table
 l1

# second example:

 l2 = mc(x,rfun = "inverse gaussian",nmb =100)$table
 l2 # yearly losses for 100 years, 365 days per year

# third example:

 l3 = mc(x,rfun = "beta", p=c(0.95))$table

# fourth example:

 l4 = mc(x,rfun = "beta",type = "binomial")$table # type of frequency distribution is chosen

# fifth example:

 l5 = mc(x,rfun = rbeta,param = list(shape1 = 0.47,shape2 = 36.66),period = "days",distname = "beta")$table
# parameters for beta distribution are given

# sixth example:

 l6 = mc(x,rfun = "normal") # comapare loss.fit.dist("normal",x) - fit is very poor

# seventh example:

 l7 = mc(x,rfun = rnorm,type = "binomial",param = list(mean=3,sd=4))$table
# parameters for normal distribution are given (and very poor, comparing loss.fit.dist("normal",x) or sixth example )

# eighth example:

 l8 = mc(x,flist= c("normal","inverse gaussian"))$table

barryrowlingson/opVaR documentation built on May 11, 2019, 7:24 p.m.