Unit-linked life insurance contracts

(an individual perspective)

Stefan Thonhauser with the aid of Josef Strini and Marco Berger

± Week 6 of Ausgangsbeschränkung

1 Introduction and overview

In this part of the lecture we are going to combine the basic models of life insurance
and financial mathematics. This happens by specifying benefits (in case of death during
the contract’s period or survival of a given maturity) which depend of the price of a
(traded risky) financial asset. At first we need to specify an appropriate model and most
importantly underlying distributional assumptions. In the following steps we discuss a
correct valuation of such contracts and the evolution of reserves. In a final step we derive
hedging strategies on an individual (per contract) basis.
Notice, here we link the benefits directly to an asset. It is also possible to link the
benefits to an account value which gets feed by returns on investment into a fund. This
is the typical situation of Fondsgebundende Lebensversicherung, whereas here we deal
with a prototypical Indexgebundende Lebensversicherung.
The following sections are based on,

• Aase & Persson, Pricing of Unit-linked Life Insurance Policies, Scandinavian Ac-
tuarial Journal, (1) 26-52, 1994;

• Møller & Steffensen, Market-Valuation Methods ind Life and Pension Insurance,
Cambridge University Press, 2007.

2 Model
At first we fix a finite time horizon T > 0 such that all actions take place in [0, T ]. All sub-
sequently introduced stochastic quantities are carried by a probability space (Ω, F, P ).
We collect all P null sets and their subsets in N . The overall flow of information is given
by a filtration {Ft }t∈[0,T ] , which fulfills the usual assumptions (right-continuous and P
completed). We use a standard Brownian motion W = (Wt )t∈[0,T ] to model economic
uncertainty and let {Gt }t∈[0,T ] denote its (completed) history.
As in the classical model we use a positive random variable Tx with intensity µ : [0, T ] →
R+ to describe the remaining life time of a x− years old individual - age x at time t = 0.

1
This life time Tx generates a filtration {Ht }t∈[0,T ] via

Ht = σ ({Tx > s | 0 ≤ s ≤ t} ∪ N ) .

One could also introduce a process I = (It )t∈[0,T ] by It = I{Tx >t} . Then, {Ht } corre-
sponds to the history of I.
In the following we will assume independence between {Gt }t∈[0,T ] and {Ht }t∈[0,T ] .
In other words, we assume that W and Tx are independent. Therefore, we assume
Ft = Gt ∨ Ht to be the total flow of information.

Remark 1. The independence assumption is crucial for the computability of explicit

values of contracts and a widely used assumption, see also the discrete time valuation
chapter of the lecture. One could also assume that the intensity µ is itself stochastics and
for example be driven by another Brownian motion W µ . If W and W µ are correlated,
then clearly Tx and W become dependent. In such situations simulation methods become
important.

Since we will focus of non-deterministic benefits, we introduce a risky asset with price
process S = (St )t∈[0,T ] . For this underlying financial asset we assume a Black-Scholes
type dynamic,

dSt = St (µ dt + σ dWt ), or equivalently,

σ2
St = S0 e(µ− 2
)t+σWt
, S0 > 0.

For discounting we need a second asset, the bank account with price St0 = er t for
t ∈ [0, T ]. For the model’s parameter we assume µ ∈ R, σ > 0, r ∈ R.
Notice, it is important that the used model for the financial market is free of arbitrage
and that we can hedge a meaningful amount of claims, ideally the market is complete.
Therefore, we could also use for example Markovian dynamics, i.e., state dependent drift
and volatility or a deterministic non-constant interest rate function.

2.1 Life insurance contracts

In the following we are interested in the market consistent value (or fair value) of an
endowment and a term insurance contract. The associated benefits are given by

bT = f (T, ST ), in case of survival of the maturity T or

bt = f (t, St ), in case of death at t ∈ (0, T ),

for some function f : [0, T ] × R+ → R+ . Of course these benefits are claims in the
financial market.
Exemplary choices for f are given in the next list:

• If f (t, St ) ≡ 1, then, we face the classical situation of deterministic benefits. Con-

sequently, the market value of it at time zero is ṽ0 = e−rt .

2
 • If f (t, St ) = St , then the market value at time zero is simply ṽ0 = S0 - one needs to
buy the asset and keep it until survival or death. In this case there is no guarantee
involved and the whole financial risk is taken by the insured.

• If f (t, St ) = max{St , K}, then there is the minimum benefit K guaranteed. Its
market value ṽ is calculated under the risk neutral measure.

• A benefit could also depend on the whole path of the asset’s price. In such a
situation bt = C(t, S[0,t] ) is a Gt measurable random variable. If we further assume
square-integrability, then we can even hedge it in the Black-Scholes model.
A nice example for such a benefit is a guarantee on the yearly return. Fix a
partition {t0 , . . . tN } of [0, T ], a base benefit K > 0 and a guaranteed return δ(ti )
for the period (ti−1 , ti ). Then, we specify
N
Sti − Sti−1
Y  
f (T, S[0,T ] ) = K max 1 + , 1 + δ(ti ) .
Sti−1
i=1

In any case, f represents a claim in the Black-Scholes model and its (financial) price
process is given by

ṽt = EQ (e−r(T −t) f | Ft ) = EQ (e−r(T −t) f | Gt ).

Where on Gt we have
( )
1 µ−r 2
 
dQ µ−r
= Zt = exp − t− Wt .
dP 2 σ σ

Remember, under Q we have dSt = St (rdt + σdWt∗ ), S0 > 0 for t ∈ (0, T ) and

ZT e−rT
 
ṽt = EP f | Gt .
Zt e−rt

3 Valuation and reserves

In this section we deal with the situation of a guaranteed benefit and look at an endow-
ment and a term insurance contract.
For allowing more flexibility we let N = (Nt )t∈[0,T ] and G = (Gt )t∈[0,T ] be two deter-
ministic functions. Here Nt denotes the number of risky assets at time t and Gt the
guaranteed value t. The benefit is described by the function

f (t, x) = Nt x ∨ Gt = max{Nt x, Gt }.

3
The function N can be thought of as an investment plan and G is a time varying
guarantee. For example we could choose

Gt ≡ G, or
Zt
Gt = πs ds, the premiums up to t, or
0
Zt
Gt = er(t−s) πs ds, the premiums with interest up to t.
0

Now we consider the pure endowment contract. If the insured is alive at time T , i.e.,
Tx > T , the benefit bT = f (T, ST ) = NT ST ∨ GT is distributed. We denote the corre-
sponding single net premium for this contract by T GX . For its determination we rely
on the assumption of risk neutrality with respect to mortality risk. This has the conse-
quence, that we use the real 
world or physicaldistribution of Tx for valuation.
Rt
Therefore we use t px = exp − µ(x + s)ds and f (t | x) = t px µx+t , where µ is esti-
0
mated from observed mortality data. In the market consistent valuation set up with the
independence assumption we have that φt = φTt φGt , where

φGt = e−rt Zt and φTt = 1.

Remark 2. Here we may cite Perrson’s PhD thesis, which presents the typical argument
for risk neutrality with respect to mortality risk. Namely, this assumption is based on
a pooling argument. I.e., an insurance company has a large number of independent and
identical contracts. From the strong law of large numbers, the aggregate number of deaths
approaches the population’s average as the number of policies gets large.This asymptotic
statement is massively present in practice, in private communications with people from
the financial market authority one often hears that the event of death can be regarded as
deterministic - one a portfolio level of course.

Remark 3. As already mentioned several times before, we distinguish in life insurance

typically between three types of calculation principles which fix mortality µ and interest
rate r.
The first order basis (r∗ , µ∗ ) is the one under which benefits and premiums are fixed
according to the equivalence principle. The second order basis (rδ , µδ ) is a decision vari-
able of the insurer , but there are legal constraints on its choice. The third order basis
(r, µ) is called the real basis and the crucial parameters for the evolution of the actual
reserve.
Generally, the first order basis is set in a conservative way. The corresponding reserve
will certainly differ from the real one, the one generated by the third order basis. An
overshoot has to be paid back in some sense to the policy holder. This can be done with
an adaption of the benefits, a premium reduction or a dividend payment.

4
Under the independence assumption, one needs to choose out of the different bases de-
pending on the purpose of the calculation. Here, for a-priori valuation of the contract
we need to use the first order basis.
Altogether, we arrive at the following representation of the net single premium - in a
market consistent way - of the endowment contract,

VG (0) = E e−rT ZT f (T, ST )1{Tx >T } = T px EQ e−rT f (T, ST ) = T px ṽ0 .

Of course the same applies for the term insurance contract,

ZT
ṼG (0) = µx+t t px EQ (e−rt f (t, St ))dt,
0

which should be derived in an exercise. For our particular benefit we get the following
explicit value.
Theorem 1. Under risk neutrality with respect to mortality and the assumptions on the
financial market, we get that for the benefit

bT = f (T, ST ) = NT ST ∨ GT

The single or net-market-consistent premium T Gx of the endowment contract is

−rT
Φ(−d02 (T )) + S0 NT Φ(d01 (T )) .
 
T Gx = T p x G T e

Here, the parameters are

Zy
1 u2
Φ(y) = √ e− 2 du,
2π
−∞

and for s ≥ t:
h i 2
Ns St
ln Gs + (r + σ2 )(s − t)
dt1 (s) = √ ,
σ s−t
h i 2
Ns St
ln Gs + (r − σ2 )(s − t)
dt2 (s) = √ .
σ s−t
Proof. We actually need to calculate just an integral, alternatively one could link the
payoff to the one of a European call and use the classical Black-Scholes Formula. Start
with,
 ∞ 
Z  
σ 2 1 w 2

T Gx = T px  NT S0 e− 2 T +σw ∨ e−rT GT √ e− 2T dw ,

2πT
−∞

5
 h h i i
1 e−rT GT σ2
and set w = σ ln NT S0 + 2 T . Then, we get

Zw Z∞
 
1 1 −w
2
− 12 ( w−σT
√ )2
T Gx = T px e−rT GT e √
dw + NT S0 p 2T e T dw
2πT (2πT )
−∞ w
    
−rT w w − σT
= T px e GT Φ √ + NT S0 Φ − √
T T
 −rT 0 0

= T px e GT Φ(−d2 (T )) + NT S0 Φ(d1 (T )) .

In the next step we focus on the corresponding reserve. In particular we analyze its
evolution, which in the classical setting is done by Thiele’s differential equation. Since
the benefits depend on the evolution of a stochastic process, the reserve will also be a
stochastic process as it will be a function of St for t ∈ (0, T ].
We consider the prospective reserve which in case that Tx > t is defined by

Vt = reserve at time t
= market value of future benefits at time t − expected value of future premiums.

This means in our situation with premium rate (πt )t∈(0,T ) , that

ZT
 

Vt = T −t px+t ṽt − e−r(u−t) πu u−t px+t du I{Tx >t} ,

t

with the option’s value

   
ṽt = EQ e−r(T −t) f (T, ST ) | Gt = EQ e−r(T −t) f (T, ST ) | St = ṽ(t, St ).

Notice, from the above theorem we have that ṽ ∈ C 1,2 (R+ × R+ ).

One way to fix the premium rate is to use an equivalence principle in the market con-
sistent framework. Namely, determine the deterministic (πt )t∈(0,T ) as solution to

ZT
T Gx = πt e−rt t px dt.
0

Then we can ensure that V0 = 0.

Theorem 2. The market value of reserves for our unit-linked endowment contract with
premiums (πt )t∈[0,T ] can be written as Vt = v(t, St )ITx >t for some function

v : [0, T ] × R+ → R+ .

6
This function v fulfills
∂ σ2 ∂ 2 ∂
v = πt + (µx+t + r)v − y 2 2 v − r y v,
∂t 2 ∂y ∂y
with the boundary condition: v(T, y) = f (T, y), at least in a viscosity sense.
Remark 4. Before we start with the proof, we need to remember Itô’s formula. Let
h ∈ C 1,2 and S = (St )t∈[0,T ] be the asset’s price with its Q dynamics,
dSt = rSt dt + σSt dWt∗ , S0 = s0 > 0.
Then, for Y = (Yt )t∈[0,T ] with Yt = h(t, St ) we have
∂ ∂ 1 ∂2
dYt = h(t, St ) dt + h(t, St ) dSt + h(t, St ) d[S]t
∂t ∂y 2 ∂y 2
∂ ∂ 1 ∂2
= h(t, St ) dt + h(t, St )(rSt dt + σSt dWt∗ ) + h(t, St ) σ 2 St2 dt
∂t ∂y 2 ∂y 2
σ2 2 ∂ 2
 
∂ ∂ ∂
= h(t, St ) + rSt h(t, St ) + St 2 h(t, St ) dt + h(t, St )σSt dWt∗ .
∂t ∂y 2 ∂y ∂y
Proof. The reserve process is given by,
ZT
 

Vt = T −t px+t ṽt − e−r(u−t) πu

 u−t px+t du I{Tx >t} ,
t
which yields that Vt is actually a function of time t, the present asset price St and It =
I{Tx >t} . Therefore we can write Vt = V (t, St )I{Tx >t} for some function V : R+ × R+ → R
and on {Tx > t}
ZT
 
e −rt
v ∗ (t, St ) := e−rt ṽt = V (t, St ) + e−r(u−t) πu u−t px+t du ,
T −t px+t
t
which is a Q martingale. The only information for the reserve process we need form the
H filtration is {Tx > t}.
e−rt ∂ ∗ ∂ ∗
We set ψ(t) = T −t px+t . For an application of Itô’s formula we need to compute ∂t v , ∂y v
2
∂
and ∂y ∗
2v .
The derivatives w.r.t. y are direct
∂ ∗ ∂ ∂2 ∗ ∂2
v = ψ(t) V (t, y), v = ψ(t) V (t, y).
∂y ∂y ∂y 2 ∂y 2
For the t derivatives we observe,
 u−t   u 
∂ ∂  Z  ∂  Z 
p
u−t x+t = exp − µ x+t+s ds = exp − µ x+s ds = µx+t u−t px+t ,
∂t ∂t   ∂t  
0 t
∂ ∂ e−rt −re−rt T −t px+t− e−rt µ x+t T −t px+t
ψ(t) = = = −(r + µx+t )ψ(t).
∂t ∂t T −t px+t (T −t px+t )2

7
 !
∂
RT RT ∂
For the premiums’ part we use ∂t g(t, u)du = −g(t, t) + ∂t g(t, u)du, to get
t t

 u  
ZT

Z
∂   
πu e−r(u−t) exp − µx+s ds du =
∂t  
t t
   u 
ZT  Zu

  Z 
− πt + πu re−r(u−t) exp − µx+s ds + πu e−r(u−t) µx+t exp − µx+s ds  du
   
t t t
Z T
= −πt + (r + µx+t ) πu e−r(u−t) u−t px+t du.
t

Such that finally,

ZT
 
d ∗
v = − (r + µx+t )ψ(t) V (t, y) + πu e−r(u−t) u−t px+t du
dt
t
ZT
 
∂
+ ψ(t)  V (t, y) − πt + (r + µx+t ) πu e−r(u−t) u−t px+t du
∂t
t
 
∂
=ψ(t) V (t, y) − (r + µx+t )V (t, y) − πt
∂t

And now for s ≥ t Itô’s formula applied to v ∗ (t, St ) results in,

Zs
∗ ∗ ∂ ∗
v (s, Ss ) = v (t, St ) + σSu v (u, Su )dWu∗
| {z } | {z } ∂y
Ys Yt t
Zs 
σ2 2 ∂ 2 ∗

∂ ∗ ∂ ∗
+ v (u, Su ) + r Su v (u, Su ) + Su 2 v (u, Su ) du
∂t ∂y 2 ∂y
t
Zs
∗ ∂
=v (t, St ) + σSu ψ(u) V (u, Su )dWu∗
∂y
t
Zs
σ2 ∂2

∂
+ ψ(u) r Su V (u, Su ) + Su2 2 V (u, Su )
∂y 2 ∂y
t

∂
+ V (u, Su ) − (r + µx+u )V (u, Su ) − πu du.
∂t

Since, v ∗ (t, St ) = EQ e−rT (ST NT ∨ GT ) | Gt , it is a martingale because of the tower

property - integrability is fine, we have already computed its value. The function ψ is

8
strictly positive, ψ(u) > 0 ∀ u ∈ (t, T ), and the choice of 0 ≤ t ≤ s ≤ T is arbitrary.
Therefore, the integrand of the du integral needs to vanish a.e., i.e.,

∂ σ2 ∂ 2 ∂
ry V (t, y) + y 2 2 V (t, y) + V (t, y) − (r + µx+u )V (t, y) − πt = 0,
∂y 2 ∂y ∂t
for allmost all (t, y) ∈ (0, T ) × R+ .

and with the associated boundary condition V (T, y) = NT s ∨ GT .

The boundary conditions are given by the behavior of f (·, ·) for S = 0 and S → ∞ - like
in the Black-Scholes model.

Remark 5. The last theorem does not say a lot on the regularity of V . But certainly it
is continuous, this suffices to show that V is a viscosity solution to the PDE. This is a
particular type of weak solution which is stronger than the notion of almost everywhere
solution and allows to show uniqueness of a solution. For us the crucial point is that we
have,
Zs
∗ ∗ ∂ ∗
v (s, Ss ) = v (t, St ) + σSu v (u, Su )dWu∗ ,
∂x
t

which gives us the martingale property.

Certainly, a C 1,2 solution to the above PDE exists, if µx+t and πt are continuous in t
and bounded.
We see that if we face the classical situation bT = f (T ), everything is independent of
the risk asset and then the PDE reduces to the common Thiele’s differential equation for
an endowment contract. On the other hand if µ and πt are set to zero, then the PDE
is exactly the Black-Scholes pricing PDE. Thus, we really derived a generalization of
Thiele’s differential equation.

As in the classical life insurance setting, we can split off the premium, πt = πtµ + πtf ,
into a mortality related and an investment related part,

πtµ = −µx+t V (t, y),

∂ σ2 ∂ 2 ∂
πtf = V (t, y) + y 2 2 V (t, y) + r y V (t, y) − r V (t, y).
∂t 2 ∂y ∂y
Remark 6. As in the Black-Scholes model we can read off a replicating strategy. Notice,
everything is conditioned on {Tx > t}. As soon as Tx realizes the reserve vanishes and
the repilcating portfolio is sold.

• In case of a single premium, V (t, St ) = T −t px+t v(t, St ) with

v(t, St ) = GT e−r(T −t) Φ(−dt2 (T )) + St NT Φ(dt1 (T )),

we can choose Ht0 = T −t px+t GT e−rT Φ(−dt2 (T )) and Ht1 = T −t px+t NT Φ(dt1 (T )).

9
 RT
• In case of a deterministic premium rate, V (t, St ) = T −t px+t v(t, St )− πu e−r(u−t) u−t px+t du.
t
−rT Φ(−dt (T ))
RT
Here we can choose Ht0 = T −t px+t GT e 2 − πn e−ru u−t px+t du and
t
Ht1 = T −t px+t NT Φ(dt1 (T )). Again we have V (t, St ) = Ht0 ert + Ht1 St .

Be always aware of the presence of St in dt1 (T ) and dt2 (T ).

It is important to state, that such a replicating strategy is at the moment only practically
motivated. If we would interpret our model as a combined financial and insurance market,
then this resulting market is incomplete. In a later chapter of this lecture we will discuss
a way to optimally hedge in such markets and meet the above strategy again.

10