Surplus process
and ruin theory
Risk theory
Warsaw University of Technology
Summer semester 2009/2010
R. Łochowski
Compound Poisson distribution
as a distribution of total claim amount
• Y - size of a (typical) claim in insurer’s
portfolio
• Y , Y1 , Y2 ,... - i.i.d. random variables
• Nt - homogeneous Poisson process with
intensity λ , modelling claims’ arrival (number
of claims between 0 and t)
• Total claim amount on time interval [0, t] may
be written as
St = Y1 + Y2 + ... + YNt
Surplus process
• Surplus process (or risk process) is a model of
accumulation of insurer’s capital
• u - the initial capital;
• c - the (constant) premium income per unit
time;
• Surplus process is defined as
Ut = u + c it − St , t ≥ 0
Moment of a ruin and ruin
probability
• When surplus process attains negative value it
means, that the claims exceed initial capital
and the income from premiums
• Moment of a ruin is defined as
T = inf {t ≥ 0 : Ut < 0}
• Ruin probability before moment t is defined
as
ψ ( u, t ) = P (T < t )
Surplus process in discrete time
• When surplus process is observed in discrete
moments t = 0,1,2,3,…, we call it a surplus
process in discrete time
Un = u + c in − Sn , n = 0,1, 2,...
• For c ≤ ES1 , from law of large numbers we get
that almost surely
limn → ∞ (Un / n ) = c − ES1 ≤ 0
• from which follows
ψ ( u ) := ψ ( u, +∞ ) = P (T < ∞ ) = 1
Loading (or safety) factor
• For c > ES1 for some θ > 0 we have
c = (1 + θ ) ES1
• The coefficient θ > 0 is called loading (or
safety) factor
• Since S1 has compound Poisson distribution,
we get c c
θ = −1 = −1
ES1 λE Y
• It is possible to prove that
ψ ( 0 ) = 1 / (1 + θ )
Adjustment coefficient
• For a positive safety factor, there exists exactly
one positive solution (in R) of the equation
e Rc = MS1 ( R ) = Ee RS1
which is called adjustment coefficient
• It may be proved that
λ + cR = λ MY ( R ) = λEe RY
Adjustment coefficient, cont.
• Problem (theoretical one): to prove statement
from the previous slide, i.e. that for the
positive loading factor, adjustment coefficient
exists and is uniquely determined
• Problem (computational): calculate the
adjustment coefficient in the special case
when claim sizes have exponential distribution
with parameter β
Lundberg inequality
• The following inequality is valid ψ ( u ) ≤ e − R iu
• Proof: let dP ( y ) = P (Y ∈ y , y + dy ) )
and ψ n ( u ) denote, that the ruin has occured
before n+1th claim and that n −1 ( )
− R iu
ψ u ≤ e
then ∞ ∞
ψ n ( u ) = ∫ ∫ ψ n −1 ( u + ct − y ) dP ( y ) λ e − λ t dt
0 0
∞ ∞
≤ ∫ ∫ exp ( −R ( u + ct − y ) ) dP ( y ) λ e − λt
dt
0 0
− R iu λ
=e ΕeY i R = e − R iu
λ + cR
Exact theoretical formula for ruin
probablility
• For every time moment t we have
− R iUt − R ( u + ct + St ) − Ru − Rct R i St
Ee = Ee =e Ee
t
=e − Ru
e − Rct
e
λ t ( MY ( R ) −1)
=e − Ru
(e − Rc − λ + λ MY ( R )
)
− Ru
=e
• and the following exact formula holds
E e − RUT
| T < ∞ iψ ( u ) = e − R iu
Functional equation for ruin
probability
• For u<0 let us define ψ ( u ) = 1
• Then we have the following formula
∞ ∞
ψ ( u ) = ∫ ∫ ψ ( u + ct − y ) dP ( y ) λ e − λ t dt
0 0
• From the above formula we derive (a bit
complicated) equation
c ∞
λ
ψ ' (u ) = ψ (u ) − ∫0
ψ ( u − y ) dP ( y )
u
= ψ (u ) − ∫
0
ψ ( u − y ) dP ( y ) − P (Y > u )
Problem (theoretical): prove the above equation
Ruin probability for exponential
claim sizes
• With dP ( y ) = β e − β y dy = p ( y ) dy ,
differentating the equation ( , ) we get
c
ψ '' ( u ) = ψ ' ( u ) − p ( 0 )ψ ( u )
λ
u
+ β ∫ ψ ( z ) p ( x − z ) dz +p ( u )
0
• Now, summing it with equation ( ) we have
c
ψ '' ( u ) + θψ ' ( u ) = 0
λ
