Nonlife Actuarial Models

Chapter 5
Ruin Theory
 Learning Objectives

1. Surplus function, premium rate and loss process

2. Probability of ultimate ruin

3. Probability of ruin before a finite time

4. Adjustment coefficient and Lundberg’s inequality

5. Poisson process and continuous-time ruin theory

2
 5.1 Discrete-Time Surplus and Ruin

• An insurance company establishes its business with a start-up capital

of u at time 0, called the initial surplus.

• It receives premiums of one unit per period at the end of each period.
Loss claim of amount Xi is paid out at the end of period i for i =
1, 2, · · ·.

• Xi are independently and identically distributed as the loss random

variable X.

• The surplus at time n with initial capital u, denoted by U (n; u), is

given by

3
 n
X
U (n; u) = u + n − Xi , for n = 1, 2, · · · . (5.1)
i=1

• The numeraire of the above equation is the amount of premium

per period, or the premium rate.All other variables are measured as
multiples of the premium rate.

• Thus, the initial surplus u may take values of 0, 1, · · · , times the

premium rate. Likewise, Xi may take values of j times the premium
rate with pf fX (j) for j = 0, 1, · · · .
2
• We denote the mean of X by μX and its variance by σX .

• We assume X is of finite support, although in notation we allow j

to run to infinity.

4
 • If we denote the premium loading by θ, then we have

1 = (1 + θ)μX , (5.2)

which implies
1
μX = . (5.3)
1+θ
We shall assume positive loading so that μX < 1.

• The business is said to be in ruin if the surplus function U (n; u)

falls to or below zero sometime after the business started, i.e., at a
point n ≥ 1.

Definition 5.1: Ruin occurs at time n if U (n; u) ≤ 0 for the first time
at n, for n ≥ 1.

5
Definition 5.2: The time-of-ruin random variable T (u) is defined as

T (u) = min {n ≥ 1 : U (n; u) ≤ 0}. (5.4)

Definition 5.3: Given an initial surplus u, the probability of ultimate

ruin, denoted by ψ(u), is

ψ(u) = Pr(T (u) < ∞). (5.5)

Definition 5.4: Given an initial surplus u, the probability of ruin by

time t, denoted by ψ(t; u), is

ψ(t; u) = Pr(T (u) ≤ t), for t = 1, 2, · · · . (5.6)

6
 5.2 Discrete-Time Ruin Theory

5.2.1 Ultimate Ruin in Discrete Time

• We now derive recursive formulas for ψ(u).

• First, for u = 0, we have

ψ(0) = fX (0)ψ(1) + SX (0). (5.7)

• Similarly, for u = 1, we have

ψ(1) = fX (0)ψ(2) + fX (1)ψ(1) + SX (1). (5.8)

• The above equations can be generalized to larger values of u as

7
 follows
u
X
ψ(u) = fX (0)ψ(u + 1) + fX (j)ψ(u + 1 − j) + SX (u), for u ≥ 1.
j=1
(5.9)

• Re-arranging equation (5.9), we obtain the following recursive for-

mula for the probability of ultimate ruin
⎡ ⎤
Xu
1 ⎣
ψ(u+1) = ψ(u) − fX (j)ψ(u + 1 − j) − SX (u)⎦ , for u ≥ 1.
fX (0) j=1
(5.10)

• To apply the above equation we need the starting value ψ(0), which
is given by the following theorem.

Theorem 5.1: For the discrete-time surplus model, ψ(0) = μX .

8
Proof: See NAM.

Example 5.1: The claim variable X has the following distribution:

fX (0) = 0.5, fX (1) = fX (2) = 0.2 and fX (3) = 0.1. Calculate the proba-
bility of ultimate ruin ψ(u) for u ≥ 0.

Solution: The survival function of X is SX (0) = 0.2 + 0.2 + 0.1 = 0.5,

SX (1) = 0.2 + 0.1 = 0.3, SX (2) = 0.1 and SX (u) = 0 for u ≥ 3. The mean
of X is

μX = (0)(0.5) + (1)(0.2) + (2)(0.2) + (3)(0.1) = 0.9,

which can also be calculated as

∞
X
μX = SX (u) = 0.5 + 0.3 + 0.1 = 0.9.
u=0

Thus, from Theorem 5.1 ψ(0) = 0.9, and from equation (5.7), ψ(1) is given

9
by
ψ(0) − SX (0) 0.9 − 0.5
ψ(1) = = = 0.8.
fX (0) 0.5
From equation (5.8), we have

ψ(1) − fX (1)ψ(1) − SX (1) 0.8 − (0.2)(0.8) − 0.3

ψ(2) = = = 0.68,
fX (0) 0.5

and applying equation (5.10) for u = 3, we have

ψ(2) − fX (1)ψ(2) − fX (2)ψ(1) − SX (2)

ψ(3) = = 0.568.
fX (0)

As SX (u) = 0 for u ≥ 3, using equation (5.10) we have, for u ≥ 4,

ψ(u) − fX (1)ψ(u) − fX (2)ψ(u − 1) − fX (3)ψ(u − 2)

ψ(u) = .
fX (0)

10
 1

0.9

0.8
Prob of ultimate ruin ψ(u)

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 5 10 15 20 25 30
Initial surplus u
5.2.2 Finite-Time Ruin in Discrete Time

• We now consider the probability of ruin at or before a finite time

point t given an initial surplus u.

• First we consider t = 1 given initial surplus u.

• As defined in equation (5.6), ψ(t; u) = Pr(T (u) ≤ t). If u = 0, the

ruin event occurs at time t = 1 when X1 ≥ 1. Thus,

ψ(1; 0) = 1 − fX (0) = SX (0). (5.20)

• Likewise, for u > 0, we have

ψ(1; u) = Pr(X1 > u) = SX (u). (5.21)

• We now consider ψ(t; u) for t ≥ 2 and u ≥ 0.

11
• The event of ruin occurring at or before time t ≥ 2 may be due to (a)
ruin at time 1, or (b) loss of j at time 1 for j = 0, 1, · · · , u, followed
by ruin occurring within the next t − 1 periods.

• When there is a loss of j at time 1, the surplus becomes u + 1 − j,

so that the probability of ruin within the next t − 1 periods is ψ(t −
˙
1; u + 1 − j ).

• Thus, we conclude that

u
X
ψ(t; u) = ψ(1; u) + fX (j)ψ(t − 1; u + 1 − j). (5.22)
j=0

Hence, ψ(t; u) can be computed as follows.

1. Construct a table with time t running down the rows for t = 1, 2, · · · ,

and u running across the columns for u = 0, 1, · · ·.

12
 2. Initialize the first row of the table for t = 1 with ψ(1; u) = SX (u).
Note that if M is the maximum loss in each period, then ψ(1; u) = 0
for u ≥ M.

3. Increase the value of t by 1 and calculate ψ(t; u) for u = 0, 1, · · ·,

using equation (5.22). Note that the computation requires the cor-
responding entry in the first row of the table, i.e., ψ(1; u), as well
as some entries in the (t − 1)th row. In particular, the u + 1 entries
ψ(t − 1; 1), · · · , ψ(t − 1; u + 1) in the (t − 1)th row are required.

4. Re-do Step 3 until the desired time point.

Example 5.3: As in Example 5.1, the claim variable X has the fol-
lowing distribution: fX (0) = 0.5, fX (1) = fX (2) = 0.2 and fX (3) = 0.1.
Calculate the probability of ruin at or before a finite time t given initial
surplus u, ψ(t; u), for u ≥ 0.

13
Solution: The results are summarized in Table 5.1 for t = 1, 2 and 3,
and u = 0, 1, · · · , 6.
Table 5.1: Results of Example 5.3

Initial surplus u
Time t 0 1 2 3 4 5 6
1 0.500 0.300 0.100 0.000 0.000 0.000 0.000
2 0.650 0.410 0.180 0.050 0.010 0.000 0.000
3 0.705 0.472 0.243 0.092 0.030 0.007 0.001
The first row of the table is SX (u). Note that ψ(1; u) = 0 for u ≥ 3, as
the maximum loss in each period is 3. For the second row, the details of
the computation is as follows. First, ψ(2; 0) is computed as
ψ(2; 0) = ψ(1; 0) + fX (0)ψ(1; 1) = 0.5 + (0.5)(0.3) = 0.65.
Similarly,
ψ(2; 1) = ψ(1; 1)+fX (0)ψ(1; 2)+fX (1)ψ(1; 1) = 0.3+(0.5)(0.1)+(0.2)(0.3) = 0.41,

14
and

ψ(2; 2) = ψ(1; 2) + fX (0)ψ(1; 3) + fX (1)ψ(1; 2) + fX (2)ψ(1; 1) = 0.18.

We use ψ(3; 3) to illustrate the computation of the third row as follows

ψ(3; 3) = ψ(1; 3) + fX (0)ψ(2; 4) + fX (1)ψ(2; 3) + fX (2)ψ(2; 2) + fX (3)ψ(2; 1)

= 0 + (0.5)(0.01) + (0.2)(0.05) + (0.2)(0.18) + (0.1)(0.41)
= 0.092.

15
 1

0.9

0.8
Prob of ruin at or before time t

0.7
Initial surplus u = 0
Initial surplus u = 5
0.6
Initial surplus u = 10
0.5

0.4

0.3

0.2

0.1

0
0 20 40 60 80 100
Time t
5.2.3 Lundberg’s inequality in Discrete Time

Definition 5.5: Suppose X is the loss random variable. The adjustment

coefficient, denoted by r∗ , is the value of r that satisfies the following
equation

E [exp {r(X − 1)}] = 1. (5.23)

Example 5.4: Assume the loss random variable X follows the distribu-
tion given in Examples 5.1 and 5.3. Calculate the adjustment coefficient
r∗ .

Solution: Equation (5.23) is set up as follows

0.5e−r + 0.2 + 0.2er + 0.1e2r = 1,

16
which is equivalent to

0.1w3 + 0.2w2 − 0.8w + 0.5 = 0,

for w = er . We solve the above equation numerically to obtain w = 1.1901,

so that r∗ = log(1.1901) = 0.1740.

Theorem 5.2 (Lundberg’s Theorem): For the discrete-time surplus

function, the probability of ultimate ruin satisfies the following inequality

ψ(u) ≤ exp(−r∗ u), (5.28)

where r∗ is the adjustment coefficient.

Proof: By induction, see NAM.

Example 5.5: Assume the loss random variable X follows the distribu-
tion given in Examples 5.1 and 5.4. Calculate the Lundberg upper bound
for the probability of ultimate ruin for u = 0, 1, 2 and 3.

17
Solution: From Example 5.4, the adjustment coefficient is r∗ = 0.1740.
The Lundberg upper bound for u = 0 is 1, and for u = 1, 2 and 3,
we have e−0.174 = 0.8403, e−(2)(0.174) = 0.7061 and e−(3)(0.174) = 0.5933,
respectively. These figures may be compared against the exact values
computed in Example 5.1, namely, 0.8, 0.68 and 0.568, respectively.

18
 5.3 Continuous-Time Surplus Function

• In a continuous-time model the insurance company receives premi-

ums continuously, while claim losses may occur at any time.

• We assume that the initial surplus of the insurance company is u

and the amount of premium received per unit time is c.

• We denote the number of claims (described as the number of occur-

rences of events) in the interval (0, t] by N(t), with claim amounts
X1 , · · · , XN(t) , which are assumed to be independently and identi-
cally distributed as X.

• We denote the aggregate losses up to (and including) time t by S(t),

19
 which is given by
N(t)
X
S(t) = Xi , (5.39)
i=1

with the convention that if N(t) = 0, S(t) = 0.

• Thus, the surplus at time t, denoted by U (t; u), is defined as

U (t; u) = u + ct − S(t). (5.40)

• Figure 5.4 illustrates an example of a realization of the surplus func-

tion U (t; u).

• To analyze the behavior of U (t; u) we make some assumptions about

the claim process S(t).

• In particular, we assume that the number of occurrences of (claim)

events up to (and including) time t, N(t), follows a Poisson process.

20
 u
Surplus U(t;u)

ruin occurred
0

Time t
Definition 5.6: N(t) is a Poisson process with parameter λ, which is
the rate of occurrences of events per unit time, if (a) in any interval (t1 , t2 ],
the number of occurrences of events, i.e., N(t2 ) − N(t1 ), has a Poisson dis-
tribution with mean λ(t2 −t1 ), and (b) over any non-overlapping intervals,
the numbers of occurrences of events are independently distributed.

• For a fixed t, N(t) is distributed as a Poisson variable with parame-

ter λt, i.e., N(t) ∼ PN (λt), and S(t) follows a compound Poisson
distribution.

• As a function of time t, S(t) is a compound Poisson process and

the corresponding surplus process U (t; u) is a compound Poisson
surplus process. We assume that the claim random variable X
has a mgf MX (r) for r ∈ [0, γ).

21
 5.4 Continuous-Time Ruin Theory

5.4.1 Lundberg’s Inequality in Continuous Time

• We first define the adjustment coefficient in continuous time. Anal-

ogous to the discrete-time case, in which the adjustment coefficient
is the solution of

1 + (1 + θ) rμX = MX (r). (5.47)

Theorem 5.3: If the surplus function follows a compound Poisson

process defined in equation (5.40), the probability of ultimate ruin given
initial surplus u, ψ(u), satisfies the inequality

ψ(u) ≤ exp(−r∗ u), (5.48)

22
where r∗ is the adjustment coefficient satisfying equation (5.47).

Example 5.6: Let U (t; u) be a compound Poisson surplus function with

X ∼ G(3, 0.5). Compute the adjustment coefficient and its approximate
value using equation (5.52), for θ = 0.1 and 0.2. Calculate the upper
bounds for the probability of ultimate ruin for u = 5 and u = 10.

Solution: The mgf of X is, from equation (2.32),

1 1
MX (r) = α
= 3
,
(1 − βr) (1 − 0.5r)
2
and its mean and variance are, respectively, μX = αβ = 1.5 and σX =
αβ 2 = 0.75. From equation (5.47), the adjustment coefficient is the solu-
tion of r in the equation
1
3
= 1 + (1 + θ)(1.5)r,
(1 − 0.5r)

23
from which we solve numerically to obtain r∗ = 0.0924 when θ = 0.1. The
upper bounds for the probability of ultimate ruin are
(
∗ 0.6300, for u = 5,
exp(−r u) =
0.3969, for u = 10.
When the loading is increased to 0.2, r∗ = 0.1718, so that the upper
bounds for the probability of ruin are
(
0.4236, for u = 5,
exp(−r∗ u) =
0.1794, for u = 10.
To compute the approximate values of r∗ , we use equation (5.52) to obtain,
for θ = 0.1,
(2)(0.1)(1.5)
r∗ ' 2 2
= 0.0864,
0.75 + (1.1) (1.5)
and, for θ = 0.2,
(2)(0.2)(1.5)
r∗ ' 2 2
= 0.1504.
0.75 + (1.2) (1.5)

24
Based on these approximate values, the upper bounds for the probability
of ultimate ruin are, for θ = 0.1,
(
0.6492, for u = 5,
exp(−r∗ u) =
0.4215, for u = 10.

and, for θ = 0.2,

(
0.4714, for u = 5,
exp(−r∗ u) =
0.2222, for u = 10.

Thus, we can see that the adjustment coefficient increases with the pre-
mium loading θ. Also, the upper bound for the probability of ultimate
ruin decreases with θ and u. We also observe that the approximation of
r∗ works reasonably well.

25