Applied Mathematical Sciences, Vol. 1, 2007, no.

38, 1851 - 1862

Convolution and Actuarial Risk

in a Pension Fund
Alvaro Tomassetti

and Sandro Tumani

Abstract
The aim of the paper is to propose a new method to evaluate the
actuarial risks in a pension fund.
We recall:
a) the utilisation of discrete Fourier transforms in the random variables (with integer realisations), to calculate convolutions in an exact
and fast way. [par 1A];
b) the denition of the contributions and benets according to the
Projected Unit Method Guidance Note 9;[par.1B].
The idea is to apply the distribution of the sum of the random variables in order to calculate directly the risk. That is done multiplying
each probability (of sum of random variables) for the corresponding
negative realisation (each realisation of any employee, for each year, is
the dierence between the accrued contributions and the corresponding
benets).
The rst application is the calculus of guarantee in a dened benet pension fund; particularly the guarantee that -in every year- the
succession of assets is not minor than the succession of liabilities.

Keywords: Exact and fast sum of discrete random variables to evaluate

the actuarial risk; Applications

1) Introduction
It is useful to give three references (the rst two, cmp[2][3]) while the third
(Ornstein-Uhlenbeck process) cmp e.g.[10].
1A) An exact and fast method to calculate the sum of random
variables with integer realisations
Let (cmp [8][11] ) X1 , X2 , ..., XN be discrete random variables (r.vs.), not
necessarily with the same distribution, having null covariance and integer realisations.
1
2

former full Professor in the University La Sapienza of Rome, Italy, [email protected]

[email protected]

1852

A. Tomassetti and S. Tumani

The sum of r.vs. is

X

(n)

n


Xs

n = 1, ..., N

(1)

s=1

For each r.v. Xs , ARs (max) is the greatest realisation and ARs (min)
the smallest realisation.
Let us denote with an integer positive pesos the number of r.vs.with the
same chacteristics. Then the maximum value of realisations is
n

M (n) =
(ARs (max) ARs (min) + 1) pesos
(2)
s=1

We write now only the conclusion of this procedure ( in order to have a

complete demonstration see ref. [11])
The nal formula [11] of the dierence between two cumulative distributions calculated by integer number (for k > m) is
FX (N) (k) FX (N) (m) =
=

2
km
+ (n)
(n)
M
M
(M (n) 1)/2

2 sin M(n) (k m)
( (n) )

M
sin
(n)
M
=1


2
(m + k 1)
cos ( (n) )
M
M (n)

Besides is ( M2
(n) ) =

2 M 1
2 0.5
n
M
s 1
s



p(s)
+
pv(s) sin(fs, )
=
v cos(fs, )
s=1

where fs, =

v=0

(3)

(4)

v=0

2 v
.
Ms

Let the sum of the arguments be


2
2
( (n) ) =
s (
)
M
M
s
s=1
n

(5)

In order to evaluate this sum we must calculate the argument of the s-th
r v and then
M
s 1
(s)
r
pv sin( 2
(s) )
M
2
) = Mr=0
tan s (
(6)
s 1

Ms
(s)
2 r
pv cos( M (s) )
r=0

1853

Convolution and actuarial risk

where r =ARs (min), ......ARs (max), are the realisations of the s-th r.v. .
) where
Then we calculate s ( 2
Ms

) .
s (
2
Ms
2

(7)

1B) Actuarial features of a defined benefit pension fund [1]

In order to realise an easy check, we write the same symbols of [1] with
very few variants.
In the pages 37-38 of [2] Blake illustrates the scheme of the dened benet
pension fund to which we refer in this paper.
Let us consider a new entry employee with age x and without any contribution year.
Let W (x, 0) denote the starting salary, T the maximum number of annual
contributions and g the projected annual growth in salary.
The salary after t years is (0 t T )
W (x, t) = W (x, 0) (1 + g)t

(8)

t
be the annual factor for service for t years of contribution,
60
P (x + t, ) the probability of remaining in the scheme between age x + t
and the old age retirement,
a
the present value of a life annuity of 1 Euro for annum,
D(x + t, ) the discount factor between age x+t and the retirement old
age , that is(1 + i)(T t) where i is the constant discount rate.
Besides is the contribution in percentage of the pensionable earning; this
contribution is prexed cmp [2], 117 ( then is not necessarily an actuarial
equilibrium rate).
After t contributions, (for an new entry of age x) the present value of
benefits, function of W(x,t) is
Let

t
W (x, t) P (x + t, ) a
(1 + i)(T t)
60

(9)

The old age pension is equal to

W (x, 0) (1 + g)t

t
60

(10)

where t represents the contribution years.

2) Definition of random variables
We want to dene the r.v. for any type of new entry in the pension Fund.

1854

A. Tomassetti and S. Tumani

2A) Introduction
(r)
We assume that the r.v. Xs (that is the s-th employee, at the end of the
r-th year) is based on two mutually exclusive events:
- the rst one is the expected value of the realisation , that is the
dierence between the expected value of the accrued contributions and the
corresponding liabilities ; this event has the probability that the s-th employee
lives until old age;
- as the only benet is the old age pension, the second event is a null
benet with probability equal to complement to 1 of the probability of the rst
event.
All the evaluations are referred to the a initial moment (r=0) by the
stochastic factor eY (r) (Ornstein-Uhlenbeck process ).
(r)
(r)
So we write Zs = Xs eY (r) .
We make the following assumptions:
(r)
a) for a xed r, the r.vs. Xs (demographic factor) are independent from
one another ;
b) the r.v. eY (r) (nancial factor) are dependent;
(r)
c) the r.v. Xs and eY (r) are mutually independent (that is the demographic factors are independent on the nancial ones).
2B) The basic idea.
The projected events of a contract are3
- at the end of the rst year the asset is the accrued capital of the
contributions paid at the beginning of the year , while the only liability of the
1
fund is the old age benet which is equal to 60
of the last annual earnings.
The pension is evaluated at the end of the year and will be paid only if the
assured person reaches the old age;
- at the end of the second year the asset will be formed by accruing the
contribution paid in the rst year (accrued for two years) plus the contribution
of the second year (accrued for one year), while the benet is the old age
2
pension which is equal to 60
of the last salary evaluated at the end of the
second year (the pension is paid only if the employee is alive at the old age),
and so on.
It is useful to underline that the problem is a succession of mutually exclusive events, that is the employee can have (at old age) the pension calculated
at the end of the first year - if only one contribution has been paid - or can
have the pension calculated at the end of the second year - if two contributions
have been paid, etc.
2C) Analysis of events
3

Pension Fund with dened benets (cmp Guidance 9 ; Retirement Benets- Actuarial
Reports of The Faculty and Institute of Actuaries) (cmp [1] , 37 )

Convolution and actuarial risk

1855

(r)

The r.v. Xs (for the s-th employee, aged x at the entry, alive at the end
of age x + r, r = 1, 2, ...T.) is formed by two mutually exclusive events:
(r)
- the rst one, is the dierence (ds ) between the expected value of the
(r)
accrued contributions((c) ds ) and the actuarial present value of the old age
(r)
r
pension calculated as 60
of the last salary( (0) ds ); to this dierence corresponds the probability to live until the old age, that is 65xr px+r ;
- the second event has probability
(1 65xr px+r ) with zero benet.
We will study the two alternatives:
I) the first event
(r)
The rst realisation ds referred at this event is formed by the dierence
between asset and liability, that is,
(r)

IA) Asset (c) ds (referred at the end of r-th year (if employee is alive)
The asset is the succession of contributions paid at the beginning of the
v-th year (v=0,1,2,...r-1) and accrued up to the end of the r-th year( r=1,2,...T)

 

r1
lx+r+1
((A))
R(0)
(1 + g) (1 + i)r ;
lx
v=0
(r)

IB) Liability (o) d s (referred at the end of the r -th year. if the employee
is alive)
It is necessary to write many specications about (cmp also [2], 37-38) :
a) x is the age of a new entry in the fund,
b) R(0) (1 + g)r the projected earning after r contributions,
r
c) 60
the accrual factor for service by r contributions,
d) (1 + g)65r the revaluation factor for earnings between age x+r and
retirement age 65,
e) a
65 the expected annuity factor (the present value of a life annuity of
Euro 1 per annum) at retirement age 65,
f) (1 + i)(65r) the discount factor between age x + r and retiremement
age 65(the discount rate is constant).
The liabilities of the pension Fund after r contributions are


lx+r+1
r
1 + g 65xr
r
)
R(0) (1 + g) (
a
65
((B))
lx
60
1+i
Then we can write that, if the assured person is alive until the end of the
r-th year, an actuarial equivalence between the accrued contributions (A) and
the actuarial present value of the corresponding liabilities (B) exists.
It is important to underline that the dierence (realisation) (A)-(B) will
exist if and only if the assured person lives until the old age (with probability

1856

A. Tomassetti and S. Tumani

65xr px+r ).

II) the second event.

The assets and liabilities will be always null if the assured person does not
live until the old age (probability 1 65xr px+r ).
A generical r.v is the following
(r)
Tab1 Random variable at the end of year ( Xs ) for the s-th
employee (aged x )
realisations
probability
lx+r+1
(R(0) ) [() ()]
65xr px+r+1
lx
0
1 65xr px+r+1
r1

(**)
(1 + g) (1 + i)r
v=0
r

(*) 60
(1 + g)r ( 1+g
)65xr a
65
1+i
If we have many r.v.s with the same
realisation and the same probabilities it is sucient to indicate only one r.v. with a repetion factor.
2D) Conclusion of definition of random variables
As actuarial equilibrium of the pension Fund, one can write about the r.v.
:
- if the sum of r.v. has all positive realisations, then the guarantee is
zero;
- if the sum of r,v. has both positive and negative realisations, then a
cost of the guarantee can exist ; this cost is the sum of the product between
negative realisations and the corresponding probabilities.
2E) Evaluation of the costs of the guarantee
(r)
According to the collective based on S employees the mean of r.v. Zs is
S

(r)
E[Zs ].
E[Z (r) ] =
s=1

We underline (for the r-th year) :

- if E[Z (r) ] = 0 then the pension fund is in actuarial equilibrium,
- if E[Z (r) ] < 0 the fund has a decit (and then the guarantee is
useful),
- if E[Z (r) ] > 0 the fund has an active ( no guarantee is needed).
3) Analysis of the actuarial risks (demographic and financial risk)
On the basis of the r.vs. (dened before) , it is easy to calculate the
distribution of the sum according to [4]
Let us denote, for the r-th year ,

1857

Convolution and actuarial risk

X (r) =

S

s=1

Z (r) =

S

s=1

(r)

Xs

and

(r)

Zs = X (r) eY (r)

From [4] we can write

(r = 1, 2, ...T. )

Demographic risk V [X (r) ] E[e2Y (r) ]

2

Financial risk E[X (r) ] V [eY (r) ]

4) Applications
Let us consider a dened benet pension fund, that provides only old age
pension at 65
The collective is based on 1000
employees (males and female with 4
dierent ages, dierent initial salaries and distribution of the employees as in
table A1)
4A) The scheme of the pension fund
The pension of old age at the moment of old age is =last salary(1 + g)t

t
where t are the contribution year (0 t 40 = T )
60
g the annual rate of variation of initial salary ( 1.50%),

gZ the annual rate of variation of initial pension ( 1.00%),

rB the annual rate of investiment in the period of benets (2.5%),
i

the annual deterministic rate (2.80%).

About the Ornstein-Uhlenbeck process(cmp [4]), let us set = 0.10 and

= 0.01,while the initial rate 0 = 0.028 and the long period rate = 0.06
If we multiply the present value of a life annuity of 1 Euro by the annual
benet , we have the actuarial present value of one pension (cmp example [2],
37-38).
The mortality table is ISTAT5 2001 in which all the death probabilities
from age 66 until 104 have been multiplied by 0.93.

Table A.1 Distribution, by age and gender, of employees and

earnings
4

We work without suscripts in X and Y in mean and variance because in this paper they
are not necessary.
5
ISTAT is the italian Organisation for statistics

1858

A. Tomassetti and S. Tumani

Males
age insured
25 100
35 260
45 120
55 120
Females
age insured
25 140
35 100
45 120
55 40

salary
1000
1100
1300
1150

term
40
30
20
10

salary
900
1000
1200
1050

term
40
30
20
10

Tax rate = 0.20 is prexed for the collective 6 .

4B) Three cases about the guarantee

It is useful to write the following three cases; we have diminuished the nal
classes of distribution from the initial fty-one to a smaller number.

Probability distribution of the sum of the r.vs. Z (1)

Table I Year = 1 Probability Distribution of the sum of the
random variables Z (1) (m = 562; = 279)
Interv.Realis
Probab
(-, 1111]
0.000 000 001 6
[-1110, -275]
0.001 392 056 6
[-274, 0]
0.004 879 249 0
[+1, +2512]* 0.993 728 692.8
Interv.( - , +2512) probability=1 000 000 000 0
* We remind that the cost is formed by only negative realisations

Table IA Year = 5 Probability Distribution of the sum of the

random variables Z (5) (m = 11844; = 1198)
6

Cmp [2] , 114

With the program trig21se.cpp one can have the distributions for the years
1,2,3,4,5,6,7,8,9,10,20,30,40
7

Convolution and actuarial risk

Interv.Realis.
(-, 19035]
[-19034, -15440]
[-15439, -13642]
[13641, -13043]
[-13042, -11844]
[-11843, -11245]
[-11244, -10646]
[-10645, -8249]
[-8248, -3455]
Interv.(-,-3455)

1859

Probab.
0.000 000 001 0
0.001 288 388 1
0.064 915 186 9
0.092 085 872 1
0.343 064 312 6
0.191 276 595 4
0.149 001 327 1
0.156 849 813 7
0.001 518 504 0
probability=1 000 000 000 0

Tab. IB Year =30 Probability Distribution of the sum of the

random variables Z (30) (m = 28812; = 1179)
Interv Realis
Probab
(-, 12449]
1010
[-12448, -6555]
1010
[-6555,+20559
0.0000000005
[+20560, +35884](*)
0.9999999995(*)
Interv.(- ,34884 probability=1 000 000 000 0
** For the costs, the positive realisation are not useful

5C) All the year

Tab A. Annual cost of guarantee in percentage of total annual
earnings . Sum of the random variables Z (r) (r=1,2.....10; 20,30,40)
r
cost
earnings cost /earnings
1
1.
1106530
< 0.0005
2
437
1121083
< 0.0005
3
2798
1135646
0.002 5
4
6729
1150174
0.005 9
5 11725 1164652
0.010 1
6 17336 1178984
.0.014 7
7 24046 1193220
0.020 2
8 30759 1207321
0.025 5
9 37911 1221288
0.031 0
10 44950 1235076
0.0364
20 50867 1165353
0.043 6
30
<1
873237
< 0.0005
40 0(n.s.) 368547
n.s.
n.s. Years with the realisations are all positive

1860

A. Tomassetti and S. Tumani

From Table A) we can observe in the rst and second year the cost/earnings
ratio is smaller than 0.0005; from the third year till the twentieth, the ratio
rises from 0.25% up to 4.36%), then from the 21-st till the 40th year the ratio
decreases.
We do not show an application about the division of the actuarial into
demographic and nancial risk, because the denition of the random variables
(cmp par 2) produces a prevalent nancial risk (for applications cmp [2]applied
to a life insurance portfolio)
6) Conclusions
This paper has the aim to present a new procedure to calculate the actuarial
risks and the costs to guarantee that the accrued contributions are sucient
to face (at the various times) the benets deriving from the contributions.
If this is not realized, it means that the accrued contributions do not even
the benets, and this causes an actuarial decit (accrued capital not sucient)
and the subject who guarantee the fund must pay this dierence.
We remind that this study considers only a dened benet pension fund,
according Guidance 9 ; Retirement Benets (Actuarial Reports of The Faculty
and Institute of Actuaries- United Kingdom)8 .
I) Methodology
The new idea is the utilisation of the distribution of the sum of the random
variables to calculate directly the cost of this guarantee.
These evaluations are made by multiplying each probability (of the sum of
the random variables) for the corresponding realisation (i.e.the only negative
dierence between accrued contributions and the corresponding benets).
It is important to underline that - for the evaluation of the cost- the differences must be negative.
II) Results about the cost of the guarantee
The collective (cmp A1) is classied into 8 homogeneous groups that distinguish themselves by the number of employees, gender, salary, age, term of the
contract (only two groups can reach 40 years of contributions - all the others
leave the fund after 10,20,30 years of contributions)
The actuarial risk of decit is negligible for the rst two years:
- then the cost (in percentage of the earnings of the whole collective)
increases from the 3-rd year ( 0,25% ) until the 20-th ( 4.36%)
- from the 21-st to the 40th year the cost diminuishes until to reach
negligible values.
If we want to analyse the division of the actuarial risk, we must consider a
homogeneous group (and not a collective classied into eight dierent groups).
8

(cmp [2] D.Blake, 2003, 37)

Convolution and actuarial risk

1861

REFERENCES
[1] D. Blake, UK pension fund management after Myners - The hunt for
correlation begins, Journal of Asset Management, 2003, 32-72.
[2] D. Blake, Pension scheme and Pension Fund in the United Kingdom,
Oxford University Press 2nd Edition, 2003.
[3] D. Blake, Pension scheme as options on pension fund assets: implications
for pension fund management, Insurance Mathematics and Economics, 1998,
263-286.
[4] M.G.Bruno, A.Tomassetti, E. Camerini, Financial and demographic
risks of a portfolio of life insurance with stochastic interest rates: Evaluation
methods and applications, North American Actuarial Journal, October 2000,
44-55.
[5] F. Cacciafesta, Discussion on M.G.Bruno, A.Tomassetti, E. Camerini,
Financial and demographic risks of a portfolio of life insurance with stochastic
interest rates: Evaluation methods and applications, North American Actuarial Journal, October 2001, 112-114.
[6] M.G.Bruno, A.Tomassetti, Authors Reply to Cacciafestas Discussion
on M.G.Bruno, A.Tomassetti, E. Camerini, Financial and demographic risks of
a portfolio of life insurance with stochastic interest rates: Evaluation methods
and applications, North American Actuarial Journal, October 2002, 110-113.
[7] M.G. Bruno, A.Tomassetti and
others, A new method for evaluating
the distribution of aggregate Claims, Applied Mathematics and Computation,
Vol 176, 2006, 488-505.
[8] M.G. Bruno, A. Tomassetti, Algorithms for sum of discrete random
variables Actuarial applicazions, WSEAS Transactions of Business and Economics, vol 3 nr. 8, 2006.
[9] M.G. Bruno, A. Tomassetti, On the calculation of convolution in actuarial applications - A case study using discrete random variables, Congress
WSEAS (Texas- USA), November 2006.
[10] G Parker, Stochastic Analysis of the Interactions between Investiments
an Insurance Risks, North American Actuarial Journal, April 1997, 55-71.

1862

A. Tomassetti and S. Tumani

[11] A. Tomassetti, A. Manna, S. Pucci, Procedimenti esatti in orizzonti

temporali niti per la valutazione della distribuzione delle varie rovine nonche
delle corrispondenti epoche nella teoria individuale del rischio. Applicazioni,
Giornale dellIstituto Italiano degli Attuari LVIII, 1995, 51-78.
Appendix nr 1
The software for the sum of discrete random variables with integer realisations (positive, negative or null) is trig20ge.cpp.
If one would like to have this programme in C++ language, one can write
an email (the rst edition of this program is twelve years old - Bibl [11], 1995)
Received: January 23, 2007