Assignment question 1

For part 1, anyone else who has an SCR of about 22.9 M with a stdev of about 300k ?
I assume he sent the same data set to everyone, mine is called 'group assignment 1A.xls'.

Rumor has it some other group has approx the same values, but they are too chicken to post it :-).

this is our confidence interval:

> ConfidenceInterval
[1] 22467877 23304912

About this first question, two things:

1) I believe it makes more sense to take a one sided confidence interval (I'd say you want to be x% sure that your SCR is high enough, not x% sure that it is between some two values, no?) I admit that it's a matter of taste...

2) Apparently you can't apply the CLT on quantiles like the VaR, i.e. the VaRs you obtained are not normally distributed.
What I believe you can do is use a binomial distribution to obtain a confidence interval. It's explained here:
http://turing.une.edu.au/~stat354/notes/node72.html

Because I'm way too kind: this is how it's implemented it in R:
calculate_conf_limit_VaR<-function(x,p,P){
x <- sort(x)
n <- length(x)
r <- max(which(pbinom((0:n),n,1-p) < (1-P)))
x[r]
}

Gil, prof. Hoedemakers heeft in de les gezegd als hint te moeten gebruiken: de CLT. Dan denken wij wel dat we mogen uitgaan van normaal verdeelde VaR

Admin schreef:About this first question, two things:

1) I believe it makes more sense to take a one sided confidence interval (I'd say you want to be x% sure that your SCR is high enough, not x% sure that it is between some two values, no?) I admit that it's a matter of taste...

2) Apparently you can't apply the CLT on quantiles like the VaR, i.e. the VaRs you obtained are not normally distributed.
What I believe you can do is use a binomial distribution to obtain a confidence interval. It's explained here:
http://turing.une.edu.au/~stat354/notes/node72.html

Because I'm way too kind: this is how it's implemented it in R:
calculate_conf_limit_VaR<-function(x,p,P){
x <- sort(x)
n <- length(x)
r <- max(which(pbinom((0:n),n,1-p) < (1-P)))
x[r]
}