Professional Documents
Culture Documents
Preliminaries ACTL30001 2023
Preliminaries ACTL30001 2023
Prof. Shuanming Li
Semester I, 2022
c Shuanming Li, 2022
Unit 0: Preliminaries
This preliminaries summarize important results, concepts, and techniques that may used in this subject,
please read this note in the first few weeks
3. Total probability formula: Suppose A 1 , A2 , . . . , An are such that Ai ∩ Aj = ∅, and ∪ni=1 Ai = Ω, then
n
X
P(B) = P(Ai )P(B|Ai),
i=1
n
X
P(B|C) = P(B|Ai , C)P(Ai |C).
i=1
4. Let X > 0 be a continuous r.v. with cdf F (x) and pdf f (x). Then
Rx
[1] F (x) = P(X ≤ x) = 0
f (y)dy [2] f (x) = F 0 (x)
R x+h
[3] P(x < X < x + h) = x
f (y)dy = F (x + h) − F (x) ≈ f (x)h, if h is very small.
R∞
[4] E[X n ] = 0
xn f (x)dx
1
5. Let X be a discrete r.v. with probability function f (x) = P(X = x), x = 0, 1, 2, . . . , and cdf F (x).
Px
[1] F (x) = y=0
f (y), x = 0, 1, 2, . . . , is a non-decreasing, right continuous, step function.
2
if Y is a continuous random variable with probability mass function g(y), y = 0, 1, 2, . . ., then
∞
X
P(X ∈ A) = E P(X ∈ A|Y ) = P(X ∈ A|Y = y)g(y),
y=0
∞
X
E[X] = E E(X|Y ) = E(X|Y = y)g(y),
y=0
Var[X] = E[Var(X|Y )] + Var[E(X|Y )]
∞ ∞
X X 2
2
= Var(X|Y = y)g(y) + E (X|Y = y) − E(X|Y = y) g(y).
y=0 y=0
Example: Let Y ∼Poisson(λ) representing the number of claims from a risk in a year. For each
claim, the probability that it is a large claim is p and the probability that it is a small claim is q = 1 − p.
Denote X to be number of large claims from this risk in a year. Then
(b) E[X] = λp
(c) Var[X] = λp
(d) X ∼Poisson(λp).
7. Let X ∼ F (x) and X ≥ 0. Then
Z ∞
E[φ(X)] = φ(0) + φ0 (x)[1 − F (x)]dx,
0
where φ(x) is a differentiable or piecewise differentiable function with E[φ(X)] < ∞. This is true
for any random variable X ≥ 0 no matter it is of continuous-type, discrete-type, or mixed-type. See
Tutorial 0 for a proof.
Remarks:
1 In particular, if φ(x) = x n , then
Z ∞
n
E[X ] = n xn−1 [1 − F (x)]dx, n > 0.
0
This is true for any positive n (not necessarily an integer).
2 If X is a discrete r.v. taking values 0, 1, 2, . . . , then
∞
X
n
E[X ] = [(k + 1)n − k n ][1 − F (k)].
k=0
3
8. Let X ∼ exp(λ1 ), Y ∼ exp(λ2 ) and X⊥Y . Then
λ1 λ2
P(X < Y ) = , P(X > Y ) = .
λ1 + λ2 λ1 + λ2
10 Some Inequalities:
e−x > 1 − x, x > 0,
− ln(1 − x) > x, 0 < x < 1,
1
x(1 − x) ≤ , 0 ≤ x ≤ 1.
4
4
11 Let X1 , X2 , . . . , Xn be independent random variables with X i following an exponential with parameter
λi . Then
Pn
(a) min(X1 , X2 , . . . , Xn ) ∼ Exp( i=1
λi ).
(b) The probability that Xj is the smallest is given by
λ
Pn j
P Xj = min(X1 , X2 , . . . , Xn ) = .
λ
i=1 i
Pn
(c) If λi 6= λj for i 6= j, then S = i=1
Xi is said to following a hyperexponential distribution with
pdf
n
X
fS (x) = Ci,n λi e−λi x , x > 0,
i=1
Qn λj
where Ci,n = j=1,j6=i λj −λi
. In particular, when n = 2,
λ1 λ2
fS (x) = λ2 e−λ2 x + λ1 e−λ1 x .
λ1 − λ2 λ2 − λ1
(d) IF λi = λj = λ, then S ∼ Γ(n, λ), this distribution is also called the Erlang(n) distribution.
(e) If λ1 , λ2 , . . . , λn may not be distinct, S is said to follow a generalised Erlang(n) distribution.
5
12. Review of continuous payment of cash-flow with rate a(t).
Let a(t) be the rate of a continuous payment at time t and A(t) be the total payment up to t. Then
(a) A(s + ds) − A(s) ≈ a(s)ds is the (approximately) total payment over the small interval (s, s + ds).
Rt
(b) A(t) = 0
a(s)ds is the total payments up to time t from time 0.
(c) dA(t) = a(t)dt.
6
14. Weighted average of a function f on [a, b].
Let f be a function on [a, b] and w(x) ≥ 0. Define
Rb
a
w(x)f (x)dx
Rb
w(x)dx
a
to be the weighted average of f on [a, b]. In particular, if w(x) = 1, then
Rb Rb
a
f (x)dx a
f (x)dx
Rb =
dx b−a
a
i=1 i=1
7
17. Change the order of double summation, integration, and product
The technique of changing the order of summation, integration and product is frequently used in this subject
to simplify calculations. Here just gives some examples.
Pb Pd Pd Pb
1. x=a y=c
f (x, y) = y=c x=a
f (x, y).
When the upper and lower bound for both the inner and outer summation are constants, just change
the order without changing the lower and upper bounds.
Pa Pa Pa Py
2. x=0 y=x
f (x, y) = y=0 x=0
f (x, y).
3.
100 200−x 200 min{y,200−y}
X X X X
f (x, y) = f (x, y)
x=0 y=x y=0 x=0
100 min{y,200−y} 200 min{y,200−y}
X X X X
= f (x, y) + f (x, y)
y=0 x=0 y=101 x=0
100 y 200 200−y
X X X X
= f (x, y) + f (x, y) .
y=0 x=0 y=101 x=0
8
4.
a y+a 2a min{x,a}
X X X X
f (x, y) = f (x, y)
y=0 x=y x=0 y=max{0,x−a}
a min{x,a} 2a min{x,a}
X X X X
= f (x, y) + f (x, y)
x=0 y=max{0,x−a} x=a+1 y=max{0,x−a}
a x 2a a
X X X X
= f (x, y) + f (x, y).
x=0 y=0 x=a+1 y=x−a
R 1 R √x R 1 R √y
5. 0 x2
f (x, y)dydx = 0 y2
f (x, y)dxdy.
6.
Z 2Z x2 Z 4Z 2
f (x, y)dydx = f (x, y)dxdy
√
1 0 0 max{1, y}
Z 1Z 2
= f (x, y)dxdy
√
0 max{1, y}
4
Z Z 2
+ f (x, y)dxdy
√
1 max{1, y}
Z 1Z 2 Z 4Z 2
= f (x, y)dxdy + f (x, y)dxdy .
√
0 2 1 y
18: Ordinary Differential Equations (ODEs)
In this part, we give a very brief introduction to linear ordinary differential equations (ODEs). These ODEs
are very important in studying Markov models.
9
Integrating factor method to solve ODE (18.1)
Rt
− a(s)ds
Let I(t) = e 0 . Clearly I 0 (t) = −a(t)I(t) with I(0) = 1. Multiplying both sides of f 0 (t) =
a(t)f (t) + b(t) by I(t) :
I(t)f 0 (t) = a(t)I(t)f (t) + I(t)b(t) ⇐⇒ I(t)f 0 (t) − a(t)I(t)f (t) = I(t)b(t)
Z t
0
⇐⇒ [I(t)f (t)] = I(t)b(t) =⇒ I(t)f (t) = I(0)f (0) + I(s)b(s)ds.
0
This in turn means that
Z t
I(0) 1
f (t) = f (0) + I(s)b(s)ds
I(t) I(t) 0
Rt Rt Z t Rs
a(s)ds a(s)ds − a(u)du
= f (0)e 0 +e 0 e 0 b(s)ds. (18.2)
0
10
Interpretation of the ODE f 0 (t) = a(t)f (t) + b(t) and its solution
• The account earns interest continuously at rate a(t), where a(t) is the force of interest at time t;
• More funds are deposited into the account in a continuouse fashion at rate b(t);
11
Interpretation of its solution in (18.3)
Recall:
Rt Z t Rt
a(s)ds a(u)du
f (t) = f (0)e 0 + e s b(s)ds.
0
1. The first term says that the amount f (0) has reached a value
Rt
a(s)ds
f (0)e 0 ;
2. the amount b(s)ds deposited in the interval (s − ds, s) has reached a value
Rt
a(u)du
e s b(s)ds;
3. the second term is the accumulated value (from time 0 to time t) of the continuous payment at rate
b(s) at time s.
12
18.2: Second order ODEs with constant coefficients
Remark: If f (t) = ert satisfies (18.4), then r has to satisfy the characteristic equation above.
13
Remarks:
• To determine k1 and k2, we need extra conditions. For example for case 1, if f (0) and f 0 (0) are
given, we can determine k1 and k2 by solving the following equations:
k1 + k2 = f (0),
k1 a1 + k2 a2 = f 0 (0).
14
Example 2: Solve the following second order ODE
f 00 (t) + 4f 0 (t) + 4f (t) = 0
with f (0) = 1 and f 0 (0) = 3.
Solution:
• f (t) satisfies the following 2nd order ODE which can be solved using the method in 18.2:
f 00 (t) = (a + d)f 0 (t) + (bc − ad)f (t).
15
19. Simulation of random variables
We want to generate values of X using random numbers generated from U (0, 1).
if u = 0.822, then X = 2
16
19.2. Simulation of a continuous r.v
If u is a random number generated from U (0, 1), then x = F−1 (u) is random number
generated from distribution F .
17
19.3. Generalized Inverse Method to Simulate of a general random variable
Remarks:
• If X is a continuous random variable with CDF F (x), then the generalised inverse is
the usual inverse, with
F −1 (u) = {x : F (x) = u}.
18
20. Hazard rate function of a continuous random variable
Let X ≥ 0 be a continuous random variable with pdf f (x), cdf F (x), and survival function S(x) = 1−F (x).
Define
f (x)
µx =
S(x)
to be the hazard rate function of X. Then R x
− µy dy
S(x) = e 0 .
21. Hazard function of a discrete random variable
Let X be a discrete random variable taking values 0 < x 1 < x2 < . . . < xk with pmf f (xj ) = P(X =
xj ), j = 1, 2, 3, . . . , and cdf F (x) = 1 − S(x) = P(X ≤ x), with F (0) = 0. Define
P(X = xj ) F (xj ) − F (xj −)
λj = P(X = xj |X ≥ xj ) = = .
P(X ≥ xj ) 1 − F (xj −)
λj (0 < λj < 1) is called the Hazard function of X at x j .
Show that
1. For j = 1, 2, . . . , k,
1 − F (xj ) S(xj )
1 − λj = = ,
1 − F (xj−1 ) S(xj−1 )
where x0 = 0.
Qr
2. j=1
(1 − λj ) = P(X > xr ) = 1 − F (xr ) = S(xr ), r = 1, 2, . . . , k.
19
22. Proper random variables and defective random variables
Remarks:
P
• If X is a proper discrete random variable with pmf f (x), then P(X < ∞) = f (x) = 1.
20