Generating random variables
Generating random variables 1 / 25
Where does “randomness” come from ?
RANDOM
NUMBER
GENERATOR
True random Pseudo-random
generator generator
Human sources Natural sources
Generating random variables 2 / 25
True random generator - Human sources
Generating random variables 3 / 25
True random generator - Natural source
Ratioactive decay:
Generating random variables 4 / 25
True random generator
Pros:
They are truly random numbers.
Cons:
Random numbers are generated slowly (compared to Pseudo-radom
generator)
Need a huge storage to record.
Systematic error.
Generating random variables 5 / 25
Pseudo-random generator (Computer algorithms)
Pseudo-random number generator (PRNG) are computer algorithms to
generate sequences of Uniform[0,1] random variables u1 , u2 , · · ·, ud .
PRNG is defined by {S, s1 , f , g} where
S is the state space (set of all possible values during the process).
s1 ∈ S is the seed.
f : S → S.
g : S → [0, 1].
Uniform random variables u1 , u2 , · · · , ud are simulated as follows:
sk = f (sk−1 ). The common algo: sk = a ∗ sk−1 + b (mod m)
uk = g(sk ) = sk /(m − 1) ∈ [0, 1]
Generating random variables 6 / 25
Pseudo-random generator
SEED
f f f f f
s1 s2 s3 ··· sd s1
g g g g g
u1 u2 u3 ··· ud u1
If seed is not specified then R initialises it using a value taken from
the system clock.
Generating random variables 7 / 25
Pseudo-random generator
Since sk ∈ {0, . . . , m − 1} it follows that after some finite number (of at
most m), a value must repeat itself and once this happens the whole
sequence will begin to repeat. In general m should be chosen to satisfy
three criteria:
1 For any initial seed, the resultant sequence has the "appearance" of
being a sequence of independent Uni[0, 1] random variables.
2 For any initial seed, the number of variables that can be generated
before repetition begins is large.
3 The values can be computed efficiently on a digital computer.
One can choose m to be a sufficiently large prime number that can be fitted
to the computer word size. For example, the ideal choice for a 32-bit word
machine is m = 231 − 1.
Generating random variables 8 / 25
Generate continuous random variables
Proposition. Let U be a uniform [0, 1] random variable. For any
continuous distribution function F , the random variable X defined by
X = F −1 (U)
has distribution F .
The inverse transform algorithm
To simulate X with distribution F :
1 Simulate a random number U in [0, 1].
2 Return X = F −1 (U) where F −1 is the inverse function of F . (Using
uniroot function in R to solve it if there is no analytical form of F −1 ).
3 Repeat the above steps n times to has n realization of X.
Generating random variables 9 / 25
Generate continuous random variables
Example 1: A continuous random variable X ≥ 0, has distribution function:
1 2
P(X ≤ x ) = 1 − −
3(x + 1) 3(x + 1)2
Is there a close form solution for X = F −1 (U) ?
Simulate a sample of n = 106 values of X.
Calculate from the sample: P(X ≤ x ) and compare them with the true
probabilities.
(If there is not a closed-form for inverse function F −1 we can use uniroot
function in R. "uniroot(f,c(a,b))$root" returns the root r of equation
f (x ) = 0 where a < r < b. Note that f (a) and f (b) must have opposite
sign).
Generating random variables 10 / 25
Generate continuous random variables
1 t 2t 2
Example 1: Closed-form solution: Let = t then 1 − − = u.
x +1 3 3
p
2 1 + 24(1 − u) − 1
→ 2t + t − 3(1 − u) = 0 → t =
4
4
→ X=p −1
1 + 24(1 − U) − 1
F.inverse<-function(u){
F.inverse<-4/(sqrt(1+24*(1-u))-1)-1}
U<-runif(10ˆ6,0,1)
X<-F.inverse(U)
Generating random variables 11 / 25
Generate continuous random variables
Example 2: The total expenses of company ABC is calculated by the sales
amount X as follow:
Total expense = x 0.1 + x 0.2 + x 0.3
where X is a random variable with distribution function
P(X ≤ x ) = 1 − 0.3e −x − 0.7e −1.5x
What is the expectation of total expense?
What is VaR90% of total expense?
Generating random variables 12 / 25
Generate continuous random variables
Example 2:
n<-10ˆ4
u<-runif(n,0,1)
x<-rep(0,n)
for (i in 1:n){
F1<-function(x){F1<-1-0.3*exp(-0.5*x)-0.7*exp(-1.5*x)-u[i]}
x[i]<-uniroot(F1,c(0,1000))$root}
totalexpense<-xˆ0.1+xˆ0.2+xˆ0.3
mean(totalexpense)
## [1] 2.749474
quantile(totalexpense,0.9)
## 90%
## 3.63548
Generating random variables 13 / 25
Generate continuous random variables
Variable Density Generate n obsers
Uniform(a, b) 1/(b − a) runif (n, a, b)
Exponential(λ) λe −λx rexp(n, λ)
Γ(a + b) a−1
Beta(a, b) x (1 − x )b−1 rbeta(n, a, b)
Γ(a) Γ(b)
β α α−1 −βx
Gamma(α, β) x e rgamma(n, α, β)
Γ(α)
(x − µ)2
!
1
Normal (µ, σ) √ exp − rnorm(n, µ, σ)
2πσ 2σ 2
Log-normal (µ, σ) Ln(X ) ∼ Normal(µ, σ) rlnorm(n, µ, σ)
α βα Install Pareto package
Pareto(α, β)
x α+1 rPareto(n, β, α)
Generating random variables 14 / 25
Generate discrete random variables
Example 1: Let X be a Bernoulli random variable where
P(X = 1) = p
P(X = 0) = 1 − p
To simulate X
1. Generate U ∼ Uniform(0,1).
2. If U ≤ (1 − p) return 0; return 1 otherwise.
p<-0.3
u<-runif(1,0,1)
x<-ifelse(u<=(1-p),0,1)
print(x)
## [1] 1
Generating random variables 15 / 25
Generate discrete random variables
Example 1: Generate a vector contains n Bernoulli random variables
p<-0.3
n<-10ˆ4
u<-runif(n,0,1)
x<-ifelse(u<=(1-p),0,1)
print(x[1:10])
## [1] 0 0 1 0 0 0 1 0 0 0
sum(x)/n #should be close to p
## [1] 0.2851
Generating random variables 16 / 25
Generate discrete random variables
Discrete random variable X has probability mass function
p0 = P(X = 0)
p1 = P(X = 1)
···
pk = P(X = k)
···
k
Let qk = P(X ≤ k) = pi with k = 0, 1, 2, · · ·; to simulate X :
P
i=0
1. Generate an uniform random variable U in [0, 1].
2. Find k such that qk−1 ≤ U < qk and return k.
Generating random variables 17 / 25
Generate discrete random variables
∞
P
with pi = 1; and
i=0
k
qk = P(X ≤ k) =
P
pi
i=0
Generating random variables 18 / 25
Generate discrete random variables
Generating random variables 19 / 25
Generate discrete random variables
Example 2: A discrete random variable X , X ≥ 0, has probability mass
function:
1 2 1
P(X = k) = p (1 − p)k + e −1
3 3 k!
where p = 0.2
Simulate a sample of N = 106 of value of X .
Calculate from the sample: P(X = 0), P(X = 1), · · · , and compare
them with the true probabilities.
WARNING: vector indices in R start with 1
Generating random variables 20 / 25
Generate discrete random variables
n<-100
p<-0.2
P0<-1/3*p+2/3*exp(-1)
P<-rep(0,n)
for (i in 1:n){P[i]<-1/3*p*(1-p)ˆi+2/3*exp(-1)/factorial(i)}
N<-10ˆ6
X<-rep(0,N)
U<-runif(N,0,1)
for (i in 1:N){
if (U[i]<P0){k<-0} else {
for (k in 1:n){if (P0+sum(P[1:k])>U[i]) {break}}}
X[i]<-k}
sum(X==2)/N
## [1] 0.165746
Generating random variables 21 / 25
Generate discrete random variables
Variables P(X = k) Generate N obsers
Bernoulli(p) pk (1 − p)(1−k) , k ∈ {0, 1} rbinom(N, 1, p)
n k
Binomial(n, p) k
p (1 − p)(n−k) rbinom(N, n, p)
Geometric(p) p (1 − p)k rgeom(N, p)
r +k−1 r
Negbinomial(r , p) k
p (1 − p)k rnbinom(N, r , p)
λk
Poisson (λ) e −λ rpois(N, λ)
k!
Generating random variables 22 / 25
Simulation exercises
Exercise 1: A random variable X has cumulative distribution function:
0 x <0
x2
0≤x <1
F (x ) = 4
x +1
4 1≤x <2
1 2≤x
2
Calculate E(e X + e X ).
Generating random variables 23 / 25
Simulation exercises
Exercise 2: Given that X1 , X2 , · · · , X5 are 5 independent Pareto random
variables with parameters:
α1 = 1.2; β1 = 2.0
α2 = 1.3; β2 = 1.8
α3 = 1.4; β3 = 1.6
α4 = 1.5; β4 = 1.4
α5 = 1.6; β5 = 1.2
Let S5 = X1 + X2 + X3 + X4 + X5 . Calculate
1 P(S5 > 10); P(S5 > 103 ); P(S5 > 105 ).
2 VaR99% (S5 ) and CVaR99% (S5 ).
Generating random variables 24 / 25
Simulation exercises
Exercise 3 (Collective model): Total claim amount of a car driver in a
year is modeled by
0
N=0
SN = N
Xi N ≥ 1
P
i=1
where N is number of claims in the year and Xi is the claim amount of the
i th accident. N and Xi are independent variables. The following assumptions
are given about the number of claim N and the claim amounts Xi :
N is a Poisson random variable with parameter λ = 2.
Xi are i.i.d Pareto random variables with parameters α = 1.5 and
β = 2.5.
Calculate P(SN > 20) and VaR99% (SN )
Generating random variables 25 / 25
