Professional Documents
Culture Documents
Yuriy Krvavych
January 2005
To my beloved wife Olya
Abstract
v
Acknowledgements
I would like to thank Professor Mike Sherris, my supervisor, for his guidance
through the early years of chaos and confusion, for his many suggestions
and constant support during this research1 . I would also like to thank my
co-supervisor Dr. Jiwook Jang who is always happy to meet up and have a
good discussion. The friendly staff in Actuarial Studies deserve many thanks
for the nice ambiance they created. Particularly want to thank Dr. Emil
Valdez and Dr. Sachi Purcal for giving their time and expertise to assist me
in completing this research. Thanks also go to Professor Zinoviy Landsman
from Haifa University for his useful advices and viewpoints.
Of course, I am grateful to my wife Olya for her patience and love.
Without her this work would never have come into existence (literally).
Finally, I am very grateful to my Lord for all his blessings and for having
given me all the opportunities and inspirations to accomplish this research.
1
The research was funded by the International Postgraduate Research Scholarship
(IPRS)
vi
Table of Contents
Abstract v
Acknowledgements vi
List of Figures x
1 Introduction 1
1.1 Reinsurance as an effective tool of insurer risk management
in shareholders value creation . . . . . . . . . . . . . . . . . 1
1.2 Motivation and structure of the thesis . . . . . . . . . . . . . 4
1.3 Declaration . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
vii
viii
6 Conclusion 207
Bibliography 212
List of Figures
2.1 Expected value and variance of the retained risk for different reinsurance
contracts R(X). The boundary line constitutes the stop-loss contracts
with r [0, ). The shaded area contains other feasible reinsurance
contracts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2 The black curve is the graph of the inverse function y = y(R) of the
solution R(y) to the problem (2.2.17). The gray line depicts the graph
y = R(y). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3 Graphical illustration of the function z(b, v) = f3 (a(b, v), b) with = 0.5.
For every fixed value v [0, 1] the function z(b, v) has its minimum on
Dv at b = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
x
xi
xii
xiii
Introduction
1
Chapter 1 2
the moral hazard and adverse selection costs of insurance risks. Self risk-
pooling arrangements are costly and insurance contracts provide an efficient
means of lowering these costs. Yet another important feature of insurance is
that unlike bondholders who can effectively reduce their credit risk exposure
by holding a well-diversified portfolio of bonds with different issuers, poli-
cyholders generally cannot mitigate insurer default risk in any cost-efficient
way. They usually accumulate their credit exposure with one or a few in-
surers, the financial strength of which is assessed by rating agencies and/or
regulators. Insurers satisfy regulatory requirements on solvency/security by
holding risk capital in addition to operating capital including a component
of premium income. Finally, risk capital is supplied by shareholders, prin-
cipals of the insurance company. By investing in the insurance company
shareholders expect from their agents (the insurance companys risk man-
agers) a fair return on their capital. On the other hand risk managers,
recognizing that companys shareholders are residual claimants, realize that
shareholders investment in the company is costly and that enhancing the
value of shareholders return is the major objective.
Efficient use of capital, such as equity capital supplied by shareholders
and insurance debt raised from selling insurance policies, by the insurance
company is a dilemma which most risk managers consider complex. In-
surers operate in the frictional environment defined by different frictional
capital costs such as costs of double taxation, costs of financial distress,
agency costs, etc. Capital efficiency requires that operational and financial
opportunities collectively result in maximum expected return subject to the
companys risk tolerance. By managing risk at the companys (enterprise)
level, the risk managers can increase the companys (shareholders) value.
One of the important tools of risk management is risk transfer. Using risk
transfer, insurers can alter their capital structure and reduce investment
Chapter 1 3
risk and mainly underwriting risk, which should lead to a decrease in ex-
pected frictional capital costs. For instance using risk transfer, insurance
risk managers can:
alter their corporate capital structure so that tax payoffs are decreased;
Insurers usually reduce the expected frictional costs through the use of rein-
surance, the classical type of risk transfer in the insurance industry.
The importance of the issue of how to manage risk efficiently in order to
maximize the shareholders value has increased during the last two decades.
Most recent industry publications (see e.g. Swiss Re Technical Publications
(1999-2004 [141]), Converuim Re Technical Publications (2001-2004 [31]),
Cologne Re Risk Insights Technical Publications (1998-2004 [30])) em-
phasize the importance of value creation as the main insurers objective and
the use of an economic value measure of insurance liabilities. Until the 1990s
tight regulation had kept competition low and profit margins high. At that
time volume was the main insurer objective. This induces selling insurance
policies at low prices to increase market share with less concern given to
underwriting risk. As a result insurance reserves were exhausted, leading to
insolvency and destroying shareholders value. Both the financial services in-
dustry and the insurance industry until recently relied heavily on accounting
methods to determine the value of insurance assets and liabilities. However,
as is now widely recognized by the industry, traditional (rule of thumb) ac-
counting methods fail to reflect the true economic situation of the insurance
company. From the economic perspective, assets and liabilities should be
Chapter 1 4
market valued taking into account all sources of insurance risks that cause
frictional costs such as insolvency costs and costs of financial distress.
It is worth noticing that the classical actuarial models of liability eval-
uation have generally ignored market risk and frictional costs that really
affect the companys value. On the other hand most financial approaches
of asset-liability evaluation in finance do take these factors into account.
There are attempts to implement some of the financial models in insurance
(e.g. RAROC for insurers). However, most financial models are unsuited for
insurers due to the peculiarities of the insurance business: insurance liabil-
ities are special type of risky debt from a financial perspective, while most
financial models are designed to evaluate financial debt. One of the possible
resolutions in this situation is to construct new effective models of risk man-
agement and economic evaluation of the shareholders value by using both
actuarial and financial approaches. This leads to the main motivations for
this dissertation.
be mentioned that this result is due to the special form of the actuarial pre-
mium principle. In fact, Borch found the optimal reinsurance strategy under
the mean-variance criterion using a mean-value premium principle for cal-
culating the premium of both the direct insurer and the reinsurer. It can be
easily shown that when the reinsurer uses the variance premium principle
then the optimal reinsurance contract is a quota share proportional rein-
surance (see eg Kaas et al (2001 [90])). Also recent results of endogenous
reinsurance optimization, obtained by Gajek and Zagrodny (2000 [59]) and
Kaluszka (2001 [92]), indicate that the minimization of the variance of an
insurance portfolio, subject to a budget constraint on the reinsurance pre-
mium, gives the optimal reinsurance contract that belongs to the class of so
called change-loss reinsurance (mixture of a quota share proportional rein-
surance and a deductible reinsurance). These results are obtained under the
assumption of a standard-deviation (dispersion) premium principle for cal-
culation of the reinsurance premium. So the reinsurance premium principle
does play a role in the matter of endogenous reinsurance optimization under
a mean-variance criterion, or in other words, the structure of the reinsurance
premium directly affects the cedents optimal decision of risk transfer from
the cedent to reinsurer. An analogous situation is observed in insurance
when we consider the problem of optimal endogenous risk sharing between
an insured and an insurer. For instance, in Moffet (1979 [105]), Raviv (1979
[123]), Spaeter and Roger (1995 [140]), and many others such as Schlesinger
(2000 [126]), Gollier (2000 [69]) the authors indicated that the form of the
Pareto optimal insurance indemnity heavily depends on the risk aversion of
insurer and the form of the cost of insurance. So in the case where the in-
surer is risk neutral and the cost of insurance is proportional to the insurance
indemnity, the optimal insurance indemnity is the deductible policy (which
is similar to the form of stop-loss risk transfer in reinsurance). However, if
Chapter 1 6
the insurer is risk averse and/or the cost of insurance is a strictly convex
function of the insurance coverage, then the optimal insurance policy is a
nontrivial deductible policy with coinsurance for losses above the deductible
(it is similar to the form of change-loss risk transfer in reinsurance). One
notes that for the most classical premium principles the expected profit net
of reinsurance and the variance of retained risk are convex functionals of
the risk transformation as a real measurable function from Hilbert space L2 .
Therefore we conclude that the problem of endogenous reinsurance optimiza-
tion under a mean-variance criterion for most actuarial forms of reinsurance
premium is equivalent to a non-linear convex optimization problem. Solving
these problems, using standard methods of convex programming, might give
a deeper insight into the theory of optimal endogenous risk transfer. This is
the motivation of Chapter Two of this thesis.
Chapter Three of this thesis is motivated by the convergence of finan-
cial and insurance markets, and the importance of hedging the corporate
risk that causes frictional costs, such as costs arisen from tax convexity
and/or costs of financial distress. From the theory and practice of corpo-
rate finance we know (see Doherty (2000 [48]) or Culp (2002 [34])) that
risk is costly to companies because it causes a set of frictional costs and
thereby decreases shareholders (corporate) value. Insurers can transfer the
risk using reinsurance to reduce the expected value of frictional costs, i.e.
to increase the shareholders value. The idea of Chapter Three is to inves-
tigate the demand for reinsurance in single-period models of maximization
of shareholders value under solvency requirements and in the presence of
frictional costs such as corporate tax and costs of financial distress. The
notion of financial distress was initially introduced in finance (e.g. see paper
by Diamond (1991 [46]), where the model of an illiquid but solvent firm is
considered, and the paper by Froot et al. (1993 [58]), where the model of
Chapter 1 7
a firm with low cash-flow is considered; also see the paper by Briys and
de Varenne (1997 [23]) and Jarrow and Purnanandam (2004 [88])). These
papers model the costs of financial distress for financial corporations with
risk-free debt, usually determined by the face value of zero-coupon bonds.
Chapter Three of this thesis introduces modelling of the financial distress
costs in an insurance company as a corporation with risky debt, and inves-
tigates the existence of risk management incentives in insurance business in
the presence of financial distress costs.
Chapter Four of this thesis is motivated by the concern about the ef-
fectiveness of different methods of reinsurance optimization in sharehold-
ers value creation in a dynamic setting. We know from de Finetti (1957
[44]) that the insurers long-run objective is to find the dividend-payment
policy (dividend cash flow) which maximizes the expected discounted sum
of all future dividend payments. This form of insurers long-run objective
has actually opened the door for the application of modern stochastic con-
trol theory to problems in insurance. In the following research papers by
Hjgaard and Taksar (1999 [82]; 2001 [83]), Asmussen et al (2000 [9]) and
Choulli et al (2003 [28]) the authors considered dividend optimization-risk
control models of the insurers liquid risk process, modelled by a Brown-
ian motion, with two control variables: dividend-payments and reinsurance
policies. They showed that there may be a demand for reinsurance in max-
imizing expected value of discounted dividend-payments paid until time of
ruin. It is worth noticing that just maximization of the expected value of
discounted dividend-payments until the time of ruin is indeed important, but
it is not the complete analysis of shareholders value creation. This is be-
cause the expected present value of future dividends (shareholders value) is
calculated under a physical probability measure, which fails to capture risk
factors, and the exogenous risk-free discount rate. And the shareholders
Chapter 1 8
1.3 Declaration
This thesis is my own original work, however includes material developed
as part of this Ph.D. thesis which has now appeared as research working
papers of which I am a joint author (these papers are downloadable from
www.geocities.org/krvavych or www.actuary.unsw.edu.au). My contri-
bution consists of deriving and analyzing the theoretical results, and per-
forming numerical examples. Chapters Two, Four and Five have not yet
appeared as research working papers.
Chapter 1 9
10
Chapter 2 11
2.1 Introduction
The problem of reinsurance optimization under a mean-variance criterion is
very classical in insurer risk management. From the point of view of agency
theory the main insurers objective is to maximize shareholders value (ex-
pected profit) subject to the risk constraints imposed by a regulator. The
idea of measuring insurance risk by variance originates from de Finetti (1940
[43]), who optimized quota share proportional reinsurance under a mean-
variance criterion. This idea is also similar to the one used in the theory
of portfolio optimization of Markowitz (1959 [100]). In the portfolio opti-
mization of Markowitz, investors maximize the expected investment return,
subject to a predetermined level of variance of the investment return, by
variables which are defined by retention levels (holding weights) of different
securities that determine the investment portfolio. In insurance, we maxi-
mize the insurers expected profit, subject to a constant on the variance, by
Chapter 2 12
optimizing the retention level of the reinsurance contract. These two opti-
mization problems have similar dual problems defined by the minimization
of variance (level of risk) subject to a predetermined level of expected profit.
Using a mean-variance criterion it is possible to optimize a reinsurance
contract, or find the optimal demand for reinsurance in two different way:
1) assume a specific form of reinsurance contract (proportional or excess-of-
loss) and find the optimal retention; or
2) consider the ceded part of the risk as a transformation of the whole risk
by a real value function and then find the optimal transformation of the risk.
In the first case the optimal reinsurance contract is exogenously prede-
termined, and in the second case the form of optimal reinsurance contract
is endogenous.
In the classical actuarial literature the most famous result in the analy-
sis of an optimal reinsurance contract of exogenous form is the one obtained
by de Finetti (1940 [43]). He found the optimal quota share proportional
reinsurance contract (optimal cedents retention level of quota share pro-
portional reinsurance) under which the variance of the portfolio achieves
its minimum when the level of expected profit is fixed (for details see also
Buhlmann (1970 [25])).
On the other hand the key result in the analysis of optimal reinsurance
contract of endogenous form was developed by Borch (1960 [14]) who con-
sidered the problem of minimizing the variance of the total claims borne by
the ceding insurer. Adopting this variance as a measure of risk, he considers
the most efficient reinsurance strategy as the one which serves to minimize
this variance. If X represents the amount of total claims with distribution
function F (X), he considers a reinsurance scheme as a transformation of
F (X). Under fairly restricted conditions he proved that the stop-loss con-
tract is most efficient in this respect. Later Kanh (1961 [91]) reconsidered
Chapter 2 13
W (X) = inf E (X ),
where belongs to the class of functions which are non-negative, convex and
equal to zero at t = 0.
Borchs result is classical in the actuarial literature (see Bowers et al
(1997 [20])), however it should be mentioned that this result is due to the
special form of the actuarial premium principle. In fact, Borch, Kahn and
Ohlin found the optimal reinsurance strategy under the mean-variance cri-
terion using a mean-value premium principle for calculating the premium
of both the direct insurer and the reinsurer. It can be easily shown that
when the reinsurer uses the variance premium principle then the optimal
reinsurance contract is quota share proportional reinsurance (see eg Kaas
et al (2001 [90])). Also recent results of endogenous reinsurance optimiza-
tion, obtained by Gajek and Zagrodny (2000 [59]) and Kaluszka (2001 [92]),
indicate that the minimization of the variance of an insurance portfolio,
subject to a budget constraint on the reinsurance premium, gives the opti-
mal reinsurance contract that belongs to the class of so called change-loss
reinsurance (mixture of a quota share proportional reinsurance and a de-
ductible reinsurance). These results are obtained under the assumption of
a standard-deviation (dispersion) premium principle for calculation of the
reinsurance premium.
We conclude that the reinsurance premium principle does matter in the
Chapter 2 14
p
2) standard deviation principle P[R(Y )] = E[R(Y )] + Var[R(Y )] with
risk loading
1
3) exponential premium principle P[R(Y )] =
ln E[eR(Y ) ], > 0 with risk
loading
1
CEX P (R(Y )) = ln E[eR(Y ) ] R(Y );
E[R(Y )eR(Y ) ]
4) Esscher premium principle P[R(Y )] = E[eR(Y ) ]
, > 0 with risk
loading
R(Y )eR(Y )
CESS (R(Y )) = R(Y ).
E[eR(Y ) ]
It is easy to check that the mean value of the costs of insurance given in 1) -
4) are positive functionals for all admissible reinsurance contracts R. Indeed,
it is obviously true for 1) and 2). For the Esscher premium principle, for all
R A, E [CESS (R(Y ))] = 0 when = 0, and
d E[R2 (Y )eR(Y ) ]E[eR(Y ) ] E2 [R(Y )eR(Y ) ]
E [CESS (R(Y ))] = =
d E2 [eR(Y ) ]
(2.2.1)
2 R(Y )
R(Y )
2
E[R (Y )e ] E[R(Y )e ]
= R(Y )
R(Y )
> 0, > 0 (2.2.2)
E[e ] E[e ]
where the latter inequality holds due to the fact that the variance of the Es-
scher transformation of R(Y ) is always positive. Therefore, E [CESS (R(Y ))]
is an increasing positive function of > 0.
For the exponential premium principle we have for all R A
E [CEX P (R(Y ))] = 0 when = 0, and
d
E [CEX P (R(Y ))] = E [CESS (R(Y ))] > 0, > 0.
d
and that P[c] = c for all constants c R. It follows from the latter property
that C(c) = 0.
Earlier we also mentioned that solving the endogenous reinsurance opti-
mization problem for the generalized dispersion reinsurance premium prin-
ciple is possible when we apply methods of non-linear convex optimization.
These methods are based on the notions of convexity and Gateaux differ-
entiability of a functional, and on the Kuhn-Tucker theorem of non-linear
convex programming (generalized Lagrange multiplier theorem). We include
some classical key definitions and results from convex analysis (see Ioffe &
Tikhomirov (1974 [86]) or Peressini et al. (1988 [118]), and Deprez and
Gerber (1985 [45])), which will be used in this chapter.
Convexity
or, taking into account this fact that the real function is non-decreasing
and convex,
The latter inequality holds and it follows from the following sequence of
rearrangements
Proof. Consider 0 < t1 < t2 and suppose that F[t2 R] < . Applying
t1
convexity and invariance-under-translation properties of F with p1 = t2
and
p2 = 1 p1 we obtain the following inequality
t1
F[t1 R] = F[p1 t2 R + p2 0] p1 F[t2 R] + p2 F[0] = F[t2 R], (2.2.5)
t2
and
R1 (Y ) R2 (Y ) F[R1 (Y )] F[R2 (Y )]. (2.2.8)
For instance, for the variance and the standard deviation principles the
property (2.2.7) is violated.
This lemma may serve as a tool for checking the necessary condition of
convexity. As an illustration we can consider the Esscher premium principle
of parameter > 0
E R(Y )eR(Y )
P[R(Y )] = .
E [eR(Y ) ]
Using this criterion we can show that the functional F[R] = D[R(Y )] is
convex. Indeed, its corresponding real function hD (t) = D[R(Y ) + tH(Y )]
for all R, U, H = U R A equals
p
hD (t) = t2 D2 [H(Y )] + 2tCov[R(Y ), H(Y )] + D2 [R(Y )]
= at2 + 2bt + c, with ac b2 0.
Gateaux differentiability
for any H R.
and
Cov[R(Y ), H(Y )]
E[H(Y )] + .
D[R(Y )]
3) Exponential principle:
1
PEX P [R] = ln E eR(Y )
Gateaux derivative of PEX P at R A is equal to
1 (R(Y )+tH(Y )) 0 E H(Y )eR(Y )
OR (PEX P ) [H] = ln E e = .
t=0 E [eR(Y ) ]
4) Esscher principle:
E R(Y )eR(Y )
PESS [R] = .
E [eR(Y ) ]
Chapter 2 25
It should be noted that PESS [R] = OR (PEX P ) [R] and Gateaux derivative
of PESS at R A is equal to
!0
E (R(Y ) + tH(Y ))e(R(Y )+tH(Y ))
OR (PESS ) [H] =
E [e(R(Y )+tH(Y )) ]
t=0
R(Y )
R(Y ) R(Y )
E H(Y )(1 + R(Y ))e E e E R(Y )e E H(Y )eR(Y )
= .
E2 [eR(Y ) ]
So the mean value, generalized dispersion, exponential and Esscher premium
principles are Gateaux differentiable on A R. However if we formally
extend their domain of definition from A to R, these premium principles
will also be Gateaux differentiable on R.
In fact there is a relationship between convexity and Gateaux differen-
tiability. This relation is introduced in the following lemma.
therefore h00 (t; R, H) 0 and by the Lemma 2.2.4 the functional F is convex.
This lemma is a useful tool for checking the convexity of a given func-
tional, especially when the form of the function h(t) is complicated. For
instance, for the exponential principle
2
E[H 2 (Y )eR(Y ) ] E[H(Y )eR(Y ) ]
h00EX P (0, R, H) = > 0, > 0
E[eR(Y ) ] E[eR(Y ) ]
Chapter 2 26
for all U, R R.
The functional F0 is called the objective functional of (2.2.9) and the func-
tional inequality F1 0 is called the inequality constraint. A point Z S
that satisfies all of the constraints of the minimization problem (2.2.9) is
called a feasible point for (2.2.9), and the set S = {Z S | F1 [Z] 0} is
called the feasibility region for (2.2.9).
If the objective functional F0 and the feasible region S are convex, then
the problem (2.2.9) is called a non-linear convex problem. It is clear that if
2
A Banach space B is reflexive iff B = B, where B is the space of linear continuous
functionals defined on B (for definition see Berezansky et al (1996 [11])).
Chapter 2 27
the functional F1 and the underlying set S are convex then the feasibility
region S is convex.
If the constraint condition of the minimization problem (2.2.9) is defined
by equality then the corresponding feasibility region S= = {Z S | F1 [Z] = 0 }
is convex if, in addition, the functional F1 is additive and homogeneous.
The following criterion for finding the optimal solution to the convex
non-linear problem is the key result in convex analysis.
F0 [Z ] = L [Z ] L [Z] = F0 [Z].
Proofs of these theorems can be found in Ioffe & Tikhomirov (1974 [86])
or Peressini et al. (1988 [118]).
where u is the cedents initial capital, P is the insurance premium income un-
derwritten in the direct insurance market, P[R(Y )] = E[R(Y ) + C(R(Y ))] is
the reinsurance premium, A is the set of all admissible reinsurance contracts
that satisfy condition 0 R(Y ) Y . The above maximization problem is
equivalent to the following minimization problem
min E[C(R(Y ))],
RA (2.2.10)
subject to Var[Y R(Y )] constant.
The authors showed that the optimal solution to such minimization problem
belongs to the class of change-loss reinsurance contracts, which includes
quota share proportional and stop-loss (deductible) reinsurance contracts.
In this subsection we will resolve the problem (2.2.11) using convex anal-
ysis under mean value, generalized dispersion, exponential premium princi-
ples and will show that under the mean value principle stop-loss reinsurance
is the unique optimal solution, whilst under the generalized premium prin-
ciple quota share proportional reinsurance is the unique optimal solution,
and under mixed mean-generalized dispersion premium principle change-loss
reinsurance (mixed proportional-nonproportional) is the optimal solution.
Let us first consider the classical method of risk theory of finding a particular
solution to the the problem (2.2.11).
and
Var[Y R(Y )] Var[Y (Y r)+ ].
and
E[Y R(Y ) r]2 E[Y (Y r)+ r]2
d d
(r) = Var[X (X r)+ ]
dr dr
Zr
d
= x2 dF (x) + r2 (1 F (r)) 2 (r)
dr
0
0
= 2 (r)(r (r)) > 0,
since
Zr
r (r) = r (1 F (r)) dx > 0.
0
Chapter 2 32
Figure 2.1: Expected value and variance of the retained risk for different reinsurance
contracts R(X). The boundary line constitutes the stop-loss contracts with r [0, ).
The shaded area contains other feasible reinsurance contracts.
d 0
= 0 = 2(r (r)) > 0, r [0, )
d
and, that
d
2
d d d (2(r (r))0 F (r)
= = =2 > 0.
d2 d 0 1 F (r)
The points ((r), (r)) are plotted for r [0, ) in Figure 2.1. It follows
from Theorem 2.2.10 that all reinsurance contracts R(), different from stop-
loss contracts, can only have an expected value and a variance of retained
risk above the curve in the , -plane, since the variance is at least as large
as for the stop-loss reinsurance with the same expected value. On the other
hand, the expected value of retained risk for reinsurance contracts R() is
not greater than the expected value of the retained risk for stop loss contract
with the same variance, i.e. stop-loss reinsurance is also a solution to the
Chapter 2 33
min E[R(Y )]
RA
Now let us reconsider this problem using the methods of convex analysis.
Consider
min Var[Y R(Y )],
RA (2.2.12)
E[R(Y )] = l < E[Y ].
For this problem we have F0 [R] = Var[Y R(Y )] and F1 [R] = E[R(Y )]
l. As it was shown above in this chapter these two functionals are con-
vex and Gateaux differentiable. The feasible region for (2.2.12) A= =
{R A | F1 [R] = 0} is convex, since the expectation is a convex homoge-
neous and additive functional. So, we deal with convex minimization prob-
lem. Therefore, according to the gradient form of the Kuhn-Tucker theorem
2.2.9 one of the necessary condition for R to be a solution to the problem
(2.2.12) is
OR (L )[H] = 0, H A=
that the optimal solution should have the following form R (y) = (y d)+ .
One can find the Lagrange multiplier using the boundary condition of the
problem (2.2.12). Indeed, for any l (0, E[Y ]) there exists d(l) > 0 such
that
E[(Y d(l))+ ] = l.
L [R] L [R ] OR (L )[R R ]
and
To find the possible form of this solution let us calculate the Gateaux
derivative of the functional L H A
Cov[R(Y ), H(Y )] 0
OR (L )[H] = (D[R(Y )]) 2Cov[Y R(Y ), H(Y )]
D[R(Y )]
= c Cov[R(Y ), H(Y )] 2Cov[Y R(Y ), H(Y )]
0 (D[R(Y )]) 1
where c = D[R(Y )]
> 0 (e.g. (x) = x : c = D[R(Y )]
; (x) = x2 : c =
2).
From here we obtain the general form of optimal (global) solution R
2 2
R(y) = y E[Y ] E[R(Y )] = ay + b,
2 + c 2 + c
and
(
0, y < d1 ;
R1 (y) =
ay + b, y d1 ,
|b|
where b < 0 and d1 = a
, are local solutions to the problem (2.2.13). Indeed,
1 a
for = 2 1a
c > 0 and Ri (y) = (ay + b)1{ydi } = a (y di )+ , (d0 =
0 iff b = 0) the following two conditions hold
Chapter 2 38
1) F1 [Ri (Y )] = 0
2) L [R] L [Ri ], R A , since according to Lemma 2.2.7
L [R] L [R ] OR (L )[R R ]
with respect to a.
Chapter 2 39
Remark 2.2.2. In the case when C is considered under the mixed mean-
generalized dispersion premium principle, i.e.
1
For exponential premium principle we have E[CEX P (R(Y ))] =
ln E[eR(Y ) ]
E[R(Y )], and we solve the following optimization problems
min E[CEX P (R(Y ))],
RA (2.2.17)
subject to Var[Y R(Y )] v < Var[Y ].
and
min Var[Y R(Y )],
RA (2.2.18)
subject to E[CEX P (R(Y ))] l < E[CEX P (Y )].
Both problems (2.2.17) and (2.2.18) are convex since the functional E[CEX P (R(Y ))]
is convex and corresponding feasible regions for these problems are convex.
Chapter 2 40
In addition we know that the functionals E[CEX P (R(Y ))] and Var[Y R(Y )]
are Gateaux differentiable, and therefore we will use a gradient form of the
Karush-Kuhn-Tucker minimization criterion (Theorem 2.2.9) to solve these
problems. First let us consider the problem (2.2.17). For this problem the
corresponding Lagrangian is equal to
eR(y) 2E eR(Y ) (y R(y)) = E eR(Y ) 2E eR(Y ) (Y R(Y )) ,
i.e. eR(y) 1 (y R(y)) = c, y 0, where 1 = 2E eR(Y ) 0,
c R is a constant. Using the natural boundary condition for the function
R : R(0) = 0 we obtain the constant c = 1. Note that 6= 0, otherwise
eR(y) = 1, or equivalently R(y) 0, but R = 0 does not belong to the
feasible range A = {R A | Var[Y R(Y )] v < Var[Y ]}.
Therefore,
1 R(y)
y= e 1 + R(y), y 0. (2.2.19)
1
From here we can see that > 0, otherwise under its negative value R(y) > y
and thus R does not belong to the feasible region. Also, 6= , otherwise
R(y) = y, but it is not a solution to (2.2.17), since3 there exists k (0, 1)
3
Here we have used the fact that E[CEX P() (Rk (Y ))] = k E[CEX P(k) (Y )], and that
E[CEX P() (Y )] is the increasing function of .
Chapter 2 41
Figure 2.2: The black curve is the graph of the inverse function y = y(R) of the solution
R(y) to the problem (2.2.17). The gray line depicts the graph y = R(y).
d d2 2
Note that dR
y(R) = 1
eR + 1 > 1 and dR2
y(R) = 1
eR > 0, and thus
y(R) is increasing convex function such that y(R) > R (see Figure 2.2).
Therefore, we conclude that there exists an inverse function R(y; ) A for
every (0, 1).
From here, using the slackness condition (Var[Y R(Y )] v) = 0, we
can find (0, 1) from the equation
Var eR(Y ; ) 1
v = Var[Y R(Y ; )] =
21
or from
2 Var eR(Y ; )
= .
4vE2 [eR(Y ; ) ]
Finally we conclude that the solution to the problem (2.2.18) exists and has
a similar analytic form.
Chapter 2 42
OZ (L )[Z] = 0, Z S= .
where
= 2Cov[Y R (Y ), V (Y )].
Chapter 2 43
So,
0.
Taking into account the following natural conditions R (0) = 0 and
y(0) = 0 we have
b 1
0 = y(0) = +c
2 2
and from here we obtain the Lagrange multiplier
1b E2 eR (Y ) E eR (Y ) + E R (Y )eR (Y )
= = > 0.
2c 2cE2 [eR (Y ) ]
Note that in general to find a necessary condition of existence of a solu-
tion to this non-convex minimization problem, might be any real number.
However, it turned out that in our particular case > 0.
Now, if in addition we assume that b + a > 0, which is equivalent to
E R (Y )eR (Y ) 2
R (Y )
< ,
E [e ]
Chapter 2 44
then
1 R (y) 1
1) y1 (R ) = 2
e (aR (y) + b) + c 2
0, since y1 (0) = y(0) = 0 and
eR (y)
y10 (R ) = (aR + b + a) > 0,
2
therefore y(R ) R ;
2) y 0 (R ) = 1 + y10 (R ) 1.
It follows that if the function R (y) is the solution to the problem (2.3.1)
E[R (Y )eR (Y ) ]
and R is such that E eR (Y ) < 2 (i.e. the reinsurance premium under
[ ]
the Esscher premium principle is less than 2 ), then R satisfies the functional
1b
relation (2.3.2) with = 2c
.
On the other hand it is assumed that the reinsurer reacts to the cedents de-
cision in the following way: it wants to find a reinsurance contract R2 () such
that the variance of ceded (assumed by the reinsurer) risk is minimal under
an assumption that the reinsurance premium income is fixed and is equal to
(1 + )E(R1 (Y )). That is, the reinsurer is trying to find the optimal reinsur-
ance contract subject to the optimal decision of the cedent. Further, based
on the reinsurers reaction, the cedent is trying to find the optimal contract
that maximizes the expected profit under the new fixed level of variance of
retained risk calculated for the reinsurer optimal reinsurance contract. This
game is repetitive until it reaches its equilibrium. This is somewhat similar
to the imperfect Cournot equilibrium in duopoly (see Varian (1992 [145])).
So the reinsurance optimal reaction to the optimal cedents decision is
such reinsurance contract R2 (Y ) that
R2 (Y ) = arg min Var[R(Y )], s.t. E[R(Y )] = E[(Y r1 )+ ] .
RA
E[(Y r2 )+ ] = E[Y (Y r1 )+ ].
E[(Y r2 )+ ],
e )] Var[Y (Y r2 )+ ],
Var[R(Y )] = Var[Y R(Y
Chapter 2 46
and thus, from the reinsurers point of view the optimal reinsurance contract
is R2 (Y ) = Y (Y r2 )+ . We can conclude from this analysis that in
equilibrium the reinsurer will accept the cedents offer if the means and
variances of the ceded part are the same under reinsurance contracts R1 (Y )
and R2 (Y )
(
E[R1 (Y )] = E[R2 (Y )],
Var[R1 (Y )] = Var[R2 (Y )]
or equivalently
(
E[(Y r1 )+ ] = E[Y (Y r2 )+ ],
.
E[((Y r1 )+ )2 ] = E[(Y (Y r2 )+ )2 ]
Note, the latter system of equilibrium equations has two trivial solu-
tions: {r1 = , r2 = 0} (no reinsurance) and {r1 = 0, r2 = } (full
reinsurance). In fact, a non-trivial equilibrium defined by finite and strictly
positive numbers r1 and r2 does not exist for all distributions. As an example
one can consider an exponential distribution of aggregate loss Y for which
there exists only trivial equilibrium. The question as to for what type of
distributions there exist a non-trivial equilibrium (a non-trivial reciprocally
optimal risk sharing) is open, and is a subject for further research.
We assume here that an insurer wants to maximize its expected profit, under
the assumption that the variance of retained risk Var[Y R(Y )] is fixed
in advance; a reinsurer determines its premium according to the standard
deviation premium principle, and thus the expected value of the cedents
retained risk equals the collected premium minus the expected value of the
retained risk minus the reinsurance premium
p
where P[Y ] is the insurance premium, PD [R(Y )] = E[R(Y )]+ Var[R(Y )]
is the reinsurance premium. If the insurer takes reinsurance in order to re-
duce the variance of its insurance portfolio, it will choose an optimal rein-
surance which is a quota share proportional reinsurance under the standard
deviation reinsurance premium principle
R1 (Y ) = a1 Y = arg min Var[R(Y )], s.t. Var[Y R(Y )] = v ,
RA
v
where a1 = 1 D[Y ]
. On the other hand the reinsurer reacts to the cedents
decision in the following way: it wants to find such reinsurance contract
R2 () under which the variance of ceded (assumed by the reinsurer) risk is
minimal under the assumption that the reinsurance premium income is fixed
and is equal to PD (R1 (Y )), and such reinsurance contract should be
R2 (Y ) = arg min Var[R(Y )], s.t. PD [R(Y )] = PD [R1 (Y )] . (2.4.1)
RA
Consider the Esscher premium principle with parameter > 0 for the rein-
surance contract R()
E R(Y )eR(Y )
PESS [R(Y ); ] = .
E [eR(Y ) ]
Chapter 2 48
P[Y ] E[Y R(Y )] PESS [R(Y ); ] = (P[Y ] E[Y ]) E [CESS [R(Y )]]
!
E R(Y )eR(Y )
= (P[Y ] E[Y ]) E[R(Y )] .
E [eR(Y ) ]
As it has been shown in (2.2.1) and (2.2.2) the mean value of cost of rein-
surance E [CESS [R(Y )]] is positive, and thus, maximization of the cedents
expected profit is equivalent to minimization of the mean value of the cost
of reinsurance. In other words the cedent will find an optimal reinsurance
contract which is a solution to the following optimization problem
min E[R(Y )e
R(Y )
]
E[eR(Y ) ]
E[R(Y )] ,
RA (2.4.2)
subject to Var[Y R(Y )] = v.
and under the assumption that insurance losses are exponentially distributed
Exp(1), and the parameter of the Esscher premium principle (0, 1). First
of all, let us determine for every v [0, 1] the domain Dv of points (a, b) on
which the equality Var[Y a(Y b)+ ] = v holds.
0 b 0
2
Z
+ [(1 a)x + ab]ex dx = 2 1 eb 2abeb + 2(1 a)2 eb
b
(1 aeb )2 = a2 2eb e2b 2a beb + eb + 1.
Chapter 2 49
Therefore,
p
b+1 (b + 1)2 (1 v)(2eb 1)
a = a(b, v) = 1.
2 eb
v 1 eb eb + 2b = f2 (b).
By comparing f1 and f2
b 2
(b + 1)2 2 eb eb + 2b e +b1
f1 (b) f2 (b) = = 0
2eb 1 2eb 1
where a < 1. It can be shown that for < 1 the minimum of the function
f3 (a(b, v), b) on Dv can be obtained when b = 0, i.e.
{b = 0} = arg min f3 (a(b, v), b) .
Dv
Chapter 2 50
Figure 2.3: Graphical illustration of the function z(b, v) = f3 (a(b, v), b) with = 0.5.
For every fixed value v [0, 1] the function z(b, v) has its minimum on Dv at b = 0.
In other words, for the cedent it is optimal to buy the quota share propor-
tional reinsurance R1 (Y ) = a1 Y with a1 = a(0, v) = 1 v. A graphical
illustration of function z(b, v) = f3 (a(b, v), b) is provided in Figure 2.3 in the
case when = 0.5. From the reinsurers side, reinsurer is interested in a
reinsurance contract R2 with minimal variance of the ceded risk under the
assumption that the reinsurance premium income is fixed and is equal to
a1 p
PESS [R1 (Y ); ] = 1a 1
= p, where p is such that 1+p = a1 = 1 v < 1
1
or equivalently p < 1
. That is
R2 (Y ) = arg min Var[R(Y )] : s.t. PESS [R(Y ); ] = p (2.4.3)
RA1
and p
aeb
PESS [R(Y ); ] = =p
(1 a) (1 a (1 eb ))
p(1 a)2
eb = ,
a(1 p(1 a))
and then by requiring eb (0, 1) we find the condition on a
p 1 p p 1
a> a> a> for p 0, .
p2 1 + p 1 + p 1
So,
pa(1 a)2 p(1 a)2
Var [a(Y b)+ ] = 2
1 p(1 a) a(1 p(1 a))
p2
1
and, in particular Var [R1 (Y )] = a21 = (1+p)2 . For every p 0, 1 the
variance Var [a(Y b)+ ] = a2 2eb e2b is a concave function of a on
p
1+p
, 1 . Therefore, in order to find the minimum value of the variance
p 1
of the ceded risk on 1+p , 1 for every fixed p 0, 1 it is enough to
p
compare the values of variance at 1+p
and 1, i.e. investigate the sign of the
function
p2 p(1 )2 p(1 )2 1
(p) = 2 on 0,
(1 + p)2 1 p(1 ) 1 p(1 ) 1
The graphs of the variance of ceded risk and the corresponding function
are provided under different values of (= 0.3; 0.5; 0.6; 0.7) in the following
figures.
= 0.3
Chapter 2 52
= 0.5
= 0.6
= 0.7
Under = 0.6; 0.7 function has different signs on its domain of defini-
tion. For instance, under = 0.7 it is negative on (0, 0.26422)(3.0691, 3.3333)
and positive on (0.26422, 3.3333). This means that only for v (0, 6.34
104 )(0.6038, 1) the optimal reinsurance contract (i.e. quota share propor-
tional reinsurance) from cedents point of view will be reciprocally optimal
for the cedent and the reinsurer.
where the risk loading E[C(R(Y ))] is a convex functional (i.e. reinsurance
premium principle is convex). It is obvious that in general [R] is a convex
functional, since the harm function h is convex and the expectation E[] is a
homogeneous and additive functional. Therefore, the optimization problem
(2.5.1) is convex and can be solved using the Karush-Kuhn-Tucker optimiza-
tion criterion 2.2.8. If in addition we assume that h differentiable, then [R]
is Gateaux differentiable, OR [H] = E[H(Y )h0 (Y R(Y ) E[Y R(Y )])],
and the problem can be solved for a particular form of h using the gradient
form of the Karush-Kuhn-Tucker criterion 2.2.9.
Here we consider the following two particular form of non-smooth harm
function h:
1) h1 (y) = y + ;
2
2) h2 (y) = (y + ) .
where the change-loss point m > 0 and the reinsurers quota share a (0, 1)
of quota share proportional reinsurance are endogenous parameters.
The proofs of these two statements can be found in the recent paper by
Gajek and Zagrodny (2004 [60]) and are constructive, and based on the idea
of using the Karush-Kuhn-Tucker optimization criterion 2.2.8.
As we can see, although the risk measures 1 and 2 of the retained risk
are quite similar we get different forms of optimal reinsurance contract. For
the first risk measure, the optimal risk reinsurance contract has the form
of stop-loss reinsurance with a contractual maximum payment 4 , while for
the second risk measure the optimal reinsurance contract is a change-loss
reinsurance. The degree of convexity of the harm function has an important
impact on the form of the optimal reinsurance contract.
the cedents retained risk is minimized subject to the fixed level of the
cedents expected profit.
where u is the cedents utility function, w = w0 + P[X].
If in addition we assume that the reinsurance premium principle is
convex and Gateaux differentiable, then taking into account the concavity
Chapter 2 57
is concave, and thus we can apply the gradient form of the Kuhn-Tucker
theorem (2.2.9) to find the necessary condition which the optimal reinsurance
contract R must satisfy. Doing so we conclude that R must satisfy the
following equality for all V
0 = OR F[V ]
This is equivalent to
Now can rewrite the necessary condition in the gradient form and get
u0 (w X [R (X)] + R (X))
GOR (X) = . (2.6.1)
E [u0 (w X [R (X)] + R (X))]
and then take the expectation and use Lemma 2.2.7 (p. 26) to obtain
Here we have also used the property of the gradient E [GOR (X)] = 1.
Summarizing we conclude that (2.6.1) is the necessary and sufficient
condition of optimality of the reinsurance contract R .
In the case where the reinsurance premium principle is mean value,
i.e. [R(X)] = (1 + )E[R(X)], the corresponding gradient is equal to the
constant 1 + , which implies that R (X) X is a constant. Taking into
account the inequality 0 R (X) X, we conclude that R = X and thus
it is optimal for the cedent not to buy reinsurance.
In some special cases of utility and reinsurance premium principle it
is possible to solve the equation (2.6.1) for R . For instance, it can be
easily shown that for the exponential utility function and the exponential
reinsurance premium principle, the optimal reinsurance contract is a quota
share proportional reinsurance with the reinsurers quota share determined
by parameters of utility function and reinsurance premium principle (see eg.
Deprez and Gerber (1985 [45])).
In the case where u is a quadratic utility function and is a general
dispersion (variance or standard deviation principles) reinsurance premium
principle both sides of the equation (2.6.1) are linear with respect to R .
Solving (2.6.1) for all admissible reinsurance contracts we can obtain that
the optimal reinsurance contract is a quota share proportional contract.
Chapter 2 59
2.7 Conclusion
This chapter has considered the problems of finding optimal endogenous
forms of reinsurance contracts under which the insurers expected profit is
maximized subject to a restriction on risk or alternatively the cedents re-
tained risk is minimized subject to a fixed level of expected profit net of
reinsurance. The approach of reinsurance optimization considered here is
different from the classical one in the actuarial literature where the form of
reinsurance is exogenously assumed (i.e. quota share proportional or excess-
of-loss). We consider the reinsurance contract as a transformation of the
underlying insurance claims, and recognize the insurers expected profit net
of reinsurance and the measure of retained risk as functionals of the rein-
surance. Using the fact that these functionals are Gateaux differentiable
and are convex for most of the classical reinsurance premium principles,
we use the methods of convex programming to solve the constrained con-
vex optimization problem with respect to the reinsurance transformation.
Solving this constrained convex problem under mean-variance criterion (i.e.
risk measure is defined by variance) we recovered the classical result that
the stop-loss reinsurance is optimal under mean-value reinsurance premium
principle, and showed that the form of optimal reinsurance is completely dif-
ferent from the stop-loss form under reinsurance premium principles other
than mean-value. It was shown that the form of optimal reinsurance depends
on the form of risk measure. The problem of finding reciprocally optimal
reinsurance endogenously in a duopoly insurer-reinsurer was considered in
Chapter 2 60
61
Chapter 3 62
distress costs.
3.1 Introduction
Often the variance of the retained total claim amount is not a satisfactory
measure of risk when an insurer is comparing different reinsurance covers.
The usefulness of the variance as a measure depends on how the relevant
random variable is spread around its mean. Positive and negative deviations
carry the same weight and both tails of the distribution together determine
the size of the variance. However, in many insurance applications only the
positive deviations (large losses) are harmful, whereas negative deviations
may even be beneficial. Moreover, after non-proportional change-loss rein-
surance the distribution of the retained risk will usually be asymmetric, and
in such a situation the usual variance as a measure of risk is not adequate.
Therefore we have to use other alternative risk measures that can prop-
erly reflect an insurers preferences. Potential candidates are ruin probability
and risk capital which is usually measured by value-at-risk (VaR) or, think-
ing coherently, by conditional value-at-risk (CVaR). Most of the classical
actuarial literature usually deals with an insurers ruin probability. The
idea to use ruin probability as a stability criterion was first considered by
Lundberg (1909 [99]) and then further developed by Cramer (1930 [32]; 1955
[33]), Buhlmann (1970 [25]), Gerber (1979 [64]) and Asmussen (2000 [7]).
In order to reduce the ruin probability (or keep it at a reasonable level)
an insurer usually takes out reinsurance cover for its insurance portfolio. The
problem of finding the optimal reinsurance that minimizes ruin probability
was first considered by Gerber (1979 [64]) and Waters (1983 [148]) employing
the idea of maximizing Lundbergs upper bound for the probability of ruin.
It was further considered by Martin-Lof (1994 [101]), Brockett and Xia (1995
Chapter 3 63
as the ratio of the expected payoff to shareholders, allowing for limited lia-
bility, to the invested risk capital. Consequently an insurers objective is to
maximize this ratio. It is worth noticing that the proposed return on risk
capital (RRC) is different from the well known risk adjusted return on
capital (RAROC, e.g. see Nakada et al (1999 [110])), which is defined as a
ratio: (risk premium plus investment return) divided by (economic capital).
So, RAROC is the type of measure of capital performance that adjusts the
returns of an insurer (or usually bank) for risk and expresses this in relation
to economic capital (risk premium plus risk capital) employed.
In this chapter, we study the demand for change-loss reinsurance contracts
in single period models M1 and M2 in the presence of corporate tax2 and
costs of financial distress.
premium income (operating capital), such that the insurers survival proba-
bility is equal to, say, (usually in practice [0.95, 0.999]). We will define
the measure of required risk capital using value-at-risk (VaR) of X in the
following way.
Definition 3.2.1. Given some confidence level (0, 1), the value-at-risk
(VaR) of a portfolio PX at the confidence level is given by the smallest
number x such that the probability that the loss X exceeds x is less than or
equal to 1 :
Therefore,
E max{0, u + P X}
(u) = 1.
u
When an insurer takes reinsurance it reduces the premium income, the vari-
ance and the value-at-risk of transformed claims (i.e. the value of umin + P
after reinsurance). The main goal of this section is to investigate whether
Chapter 3 68
where umin (a, b) and P (a, b) are corresponding values of the required
minimal risk capital and premium income after reinsurance. The first model
is conservative to some extent. It does not allow the insurer to reduce the
required minimal risk capital after purchasing reinsurance below the level
of required minimal risk capital determined at the beginning of period4 .
However, the direct insurer can change both initial risk capital and the
parameters of the change-loss reinsurance to achieve a maximum return
on risk capital. In the second model it is assumed that the insurer is allowed
to alter (reduce) the required minimal risk capital by taking reinsurance5 ,
4
In some states of the European Union, particularly in France, primary insurers are not
allowed to deduct reinsurance from their own technical reserve (risk capital) requirements.
5
In the USA and within Lloyds, for example, any (re)insurer purchasing insurance
from another firm can deduct that cover from its own capital requirement.
Chapter 3 69
and moreover, after taking reinsurance the cedent holds exactly such amount
of risk capital (umin (a, b)) that just satisfies minimum solvency requirements.
We will denote the retained risk by Ia,b (X) = X Ja,b (X).
where umin = VaR [X] (1 + )E[X]. After taking reinsurance from class
J the cedents premium income becomes
where > 0 is the reinsurers risk loading. We assume that > , i.e.
reinsurance loading is higher because it corresponds to a riskier loss. It is a
reasonable assumption since it follows from empirical arbitrage constraints
imposed by arbitrage avoidance (see Venter (1991 [147])):
1) additivity;
2) a premium calculation principle should produce a higher risk loading,
relative to expected losses, for an excess of loss cover than for a primary
cover on the same risks.
One of the principles that can meet the above constraints is the mean
value premium principle applied to an adjusted (distorted) probability dis-
tribution. In our case we have
1
where G(x) = F (kx) and k = 1+
(0, 1) is a risk adjustment coefficient.
Assuming a certain value of risk adjustment coefficient k we can properly
Chapter 3 70
determine risk loading for the reinsurer from the following equation
Z
(1 + )EF [a(X b)+ ] = EG [a(X b)+ ] = a(x b)dG(x)
b
Z Z
a
= a (1 F (kx))dx = (1 F (x))dx. (3.2.4)
k
b bk
It is obvious that the premium income P (a, b) is positive for all a [0, 1]
and b [0, ], indeed
where u1 0 is the excess of required minimal risk capital, such that total
risk capital is equal to u = umin + u1 .
It follows from this that the return on risk capital after reinsurance is
the following function of u1 , a and b
E max{0, S(umin + u1 ; a, b)}
(umin + u1 ; a, b) = 1,
VaR [X] P + u1
and our goal is to find u1 , a and b such that
We first find the distribution of retained risk Ia,b (X) in order to calculate
the risk capital and the expected value of the cedents limited liability.
Consider the case where a 6= 1.
R
where (u1 , a, b) = VaR [X]+u1 (1+(b, ))a (1F (x))dx > VaR [X]
b
P > 0.
In order to calculate the latter integral we will use a result from proba-
bility.
Lemma 3.2.2. For any random variable Z with continuous and al-
most everywhere differentiable cdf G the following equality holds
Zc Zc
c [0, ) (c z)dG(z) = G(z)dz.
0 0
(uZ1 , a, b)
and
(u1 , a, b)ab
Zb Z
1a
F (y)dy + (1 a) F (y) dy
a
0 b
(u1 , a, b) ab (u1 , a, b) ab
= (1 a) F
a 1a 1a
(u1 , a, b)ab
Z1a
(u1 , a, b) b (u1 , a, b) ab
F (y)dy = F
1a 1a
b
(u1 , a, b)ab
Z1a Z
(u1 , a, b) ab
F (y)dy (1 + (b, ))F (1 F (y))dy
1a
b b
(u1 , a, b) ab (u1 , a, b) b
F
1a 1a
(u1 , a, b)ab
Z
1a Z
F (y)dy (1 + (b, )) (1 F (y))dy
b b
(u1 , a, b)ab
Z
1a
(u1 , a, b) ab
= F (1 F (y))dy
1a
b
Chapter 3 74
Z
(1 + (b, )) (1 F (y))dy
b
(u1 , a, b)ab
Z
1a
(u1 , a, b) ab
= F (1 F (y))dy
1a
b
Z
(1 + ) (1 F (y))dy
< 0.
b
1+
The latter inequality holds, since for any > 0 b, (u1 ,1a
a, b)ab b
1+ , .
Therefore, the function ( , a, ) decreases on [0, 1). So, for every fixed
excess of required minimal risk capital u1 0 the return on risk capital
takes its maximum value on [0, 1) when a = a = 0 or equivalently when
b = b = . Moreover, the integral in (3.2.6) is a continuous function of a
on [0, 1], since
(uZ1 , a, b) (uZ1 , 1, b)
We can verify the latter equality by considering the following two cases:
1) If for some fixed b (u1 , 1, b) < b then a0 > 0 such that a > a0
(u1 , a, b) < b and
(uZ1 , a, b) (uZ1 , a, b)
Figure 3.1: Graphical illustration of excess of required minimal risk capital u1 (a, b)
under fixed level 10% of return on equity: exponential case
and we conclude that the return on risk capital attains its maximum value
R [X]
VaR R [X]
VaR
F (y)dy P (1 F (y))dy
0 0
1= >0
VaR [X] P VaR [X] P
at (u = umin ; a = 0 b = ).
Figure 3.2: Graphical illustration of excess of required minimal risk capital u1 (a, b)
under fixed level 10% of return on equity: Pareto case
3.1 and 3.2 we can see the surface of all indifference points (u1 , a, b) under
which the return on equity is the same fixed value (10%). We can see that
the less change-loss reinsurance the cedent takes (a decreases or/and b in-
creases) the more risk capital is needed to provide the predetermined fixed
return, and vice versa.
In this model the direct insurer is allowed, under a fixed solvency level, to
reduce minimal risk capital by taking into account the purchase of change-
loss reinsurance. The required minimal risk capital will then depend on the
parameters of the change-loss reinsurance.
After reinsurance the cedents surplus is equal to
In order to calculate this, first we have to find the value-at-risk of the trans-
formed retained risk after reinsurance, i.e. VaR [Ia,b (X)]. According to
Definition 3.2.1
or equivalently
(
VaR [Ia,b (X)], VaR [Ia,b (X)] < b;
VaR [X] = VaR [Ia,b (X)]ab
1a
, VaR [Ia,b (X)] b.
We then have
(
VaR [X], VaR [X] < b;
VaR [Ia,b (X)] =
ab + (1 a)VaR [X], VaR [X] b.
where
(
ab + (1 a)VaR [X], b VaR [X];
(a, b) = VaR [Ia,b (X)] =
VaR [X], b > VaR [X],
For a [0, 1) cdf FIa,b is continuous, and thus
Rb R
(a,b)
F (y)dy + F yab dy, b VaR [X]
1a
0 b
V (a, b) =
VaRR [X]
F (y)dy, b > VaR [X]
0
VaRR [X] R [X]
VaR
F (y)dy a F (y)dy, b VaR [X]
0 b
=
VaRR [X]
F (y)dy, b > VaR [X].
0
400
300
VHa,bL
200
300
100
0 200
0
b
0.25
0.5 100
a 0.75
10
Z [X]
VaR Z [X]
VaR
= F (y)dy.
0
Therefore, a [0, 1]
VaRR [X] R [X]
VaR
F (y)dy a F (y)dy, b VaR [X]
0 b
V (a, b) = (3.2.10)
VaRR [X]
F (y)dy, b > VaR [X].
0
R [X]
VaR
We see that the global maximum of V is equal to F (y)dy. Moreover,
0
if the cedents objective is to maximize the expected value of its limited lia-
bility at the end of period, then there is no demand for reinsurance contracts
Chapter 3 81
from the class of change-loss reinsurance contracts, since V (a, b) attains its
local maximum when a = 0 or/and b = VaR [X] (see Figure 3.3).
However, there may be a demand for change-loss reinsurance in the case
V (a,b)
where the cedents objective is to maximize the return (a, b) = u(a,b)
1
(or gross return 1 + (a, b)) on risk capital supplied by shareholders at the
beginning of period. First of all the required risk capital decreases on {a
[0, 1]} {b > VaR [X]} when a tends to 0 and/or b tends to . This means
that the return on equity attains its local maximum when a = 0 or b = ,
i.e.
R [X]
VaR
F (y)dy
0
max (1 + (a, b)) =
{a[0,1]}{b>VaR [X]} VaR [X] P
and thus there is no demand for reinsurance contracts from the subclass
{a [0, 1]} {b > VaR [X]}.
When the threshold of change-loss reinsurance b VaR [X], the value of
the required risk capital changes in the following way: for any fixed a [0, 1]
b
u(a, b) = aF 0, (3.2.11)
b 1+
that is u(a, ) increases on (0, VaR [X]), and on {a [0, 1]}{b > VaR [X]}
b
u(a, b) = a 1 F 0, (3.2.12)
b 1+
in this case insurance premium income and the expected value of its limited
liability at the end of the period are equal to 0, and thus the insurer is out of
business (or it is replaced by the reinsurer). To avoid this degenerate situa-
tion we restrict the reinsurers quota share in the domain of all change-loss
reinsurance contracts by an upper bound a1 < 1. This will guarantee the
existence of both the insurer and reinsurer in the market.
The main aim of this subsection is to show that in contrast to model
M1 under specific conditions there may be a demand for reinsurance in
model M2. Moreover, it is difficult to tackle the problem of maximization
of the return on the risk capital supplied by shareholders in the case of
a general form of the distribution function F of clams size. Therefore, to
provide intuition about the results we will restrict ourself to the case of
exponentially distributed claims size.
Let F (x) = 1 ex , > 0, x 0. Then x = F 1 (y) = ln(1y)
, and
thus
ln(1 )
VaR [X] = F 1 () =
1 + 1 + (b, ) b 1 + b
P (a, b) = ae = 1 ae 1+ . (3.2.14)
V (a, b) =
a b + ln(1) + 1 eb (1 ) ln(1)
, b [0, VaR [X]];
=
ln(1)
, b (VaR [X], ).
300
uHa,bL 200
100 300
0 200
0
b
0.25
0.5 100
a 0.75
10
(3.2.8) becomes
ln(1 ) 1
u(a, b) = ab (1 a) + (1 + (b, ))aeb (1 + )
ln(1 ) 1 b
1+
= a b+ + (1 + )e (3.2.15)
ln(1 ) 1 +
.
We now investigate the question as to whether there is a demand for change-
loss reinsurance in shareholders value creation at all. In order to do this
we examine the ratio 1 + (a, b) (total return on equity) in the case of
exponentially distributed cedents aggregate claims size X.
To illustrate we assume that = 0.975; a1 = 0.92 (the upper bound of
quota share); = 0.4 (the risk adjustment coefficient k = 0.7143) and
1
= 0.01 (E[X] =
= 100). For this particular case the total return on risk
V (a,b)
capital as the function 1 + (a, b) = u(a,b)
, defined on {a [0, a1 ]} {b
VaR [X]}, attains its local maximum 1.24693 at the point (a=0.92; b=95.11)
Chapter 3 84
1.225
1 + 1.2
300
1.175
1.15
0 200
b
0.2
0.4 100
0.6
a
0.8
0
Figure 3.5: Graphical illustration of the total return on equity as the function 1 +
V (a,b)
(a, b) = u(a,b) defined on {a [0, a1 ]} {b VaR [X]}
(see Figure 3.5). On the other hand the local maximum of 1 + (a, b) on
{a [0, a1 ]} {b > VaR [X]} is equal to 1.1857 for b = (no reinsurance).
So, the change-loss reinsurance contract with b = 95.11 and a = a1 = 0.92)
is an optimal contract under which the cedents return on risk capital is
maximized.
1 +
1.24
1.22
b
50 100 150 200 250 300 350
1.18
1.16
1.14
V (a1 ,b)
Graphical illustration of the total return on equity as the function 1+(a1 , b) = u(a1 ,b)
Summarizing one may conclude that there may be demand for reinsur-
ance in model M2. This means that in principle, an insurer might create
value for shareholders by altering its capital structure using reinsurance, af-
ter having issued insurance. Indeed, due to the peculiarities of the insurance
business an insurer is generally leveraged via the insurance market. That
is, its debt is raised through the sale of insurance policies rather than via
capital markets (although in addition an insurer can issue additional debt in
a capital market). Purchasing reinsurance effectively reduces the insurers
debt and risk capital required by the regulator, and thus changes the finan-
cial leverage of the insurance company. Alternatively, considering model M2
we can think of reinsurance as an additional layer of synthetic equity capi-
tal that substitutes costly risk capital and thus increases return on equity.
Therefore, the decision to reinsure can be treated as both a risk-management
and a capital-structure tool in shareholders value creation.
In contrast, model M1 is conservative since it does not allow the insurer
to reduce risk capital after taking reinsurance. Therefore, holding extra risk
capital offsets the demand for reinsurance.
Remark. In this chapter we consider the actuarial approach of insurer
risk management. This means that the gross premium P does not reflect
the effect of insolvency on policy payoff. Using an economic approach of
insurance asset-liability modelling we can redefine single period models M1
and M2 in the following way. As earlier, all premiums are collected at the
beginning of the period and all insurance claims are paid at the end of the
period. At the beginning of the period the insurers assets A0 consist of
premiums P0 and risk capital (equity) E0 supplied by shareholders. All
assets at time 0 are invested in financial instruments with time-1-payoff
A1 = (1+rA )A0 . The terminal value of insurance claims (losses) is a random
variable L1 .
Chapter 3 86
P0 = er EQ [1 ] = er EQ L1 1{A1 >L1 } + A1 1{A1 <L1 }
= er EQ L1 (L1 A1 ) 1{A1 <L1 } ,
where r is the risk-free interest rate, Q is the risk-neutral risk measure. In the
latter equality the second term represents the value of insolvency exchange
option, and we can see that the premium under new economic assumption
of limited liability is less than one calculated using an actuarial approach.
Under the fair (equilibrium) pricing the value of equity is
E0 = er EQ (A1 L1 ) 1{A1 >L1 } .
P0 = er EQ L1 (L1 (1 + rA )(P0 + E0 )) 1{(1+rA )(P0 +E0 )<L1 }
E0 = er EQ ((1 + rA )(P0 + E0 ) L1 ) 1{(1+rA )(P0 +E0 )>L1 } .
Chapter 3 87
to Pb0 and E
b0 . This new economic equilibrium model of the insurer should
not impose any demand for reinsurance in the maximization of the return
on equity, unless frictional costs such as taxes and costs of financial distress
are included.
In both models M1 and M2 the gross premium is determined using mean
value premium principle without any adjustment with respect to the value of
insolvency put. This possibly cause the situation where model M2 imposes
demand for reinsurance in frictionless environment.
VS (u, a, b) = Ei EIa,b (X) [max {(1 + i) (u + P (a, b)) Ia,b (X); 0} | i ]
Z1
= Ei (1 y) dFIa,b (y) ,
0
where
1 = (1 + i) (u + P (a, b))
Z
= (1 + i) VaR [X] + u1 (1 + (b, ))a (1 F (x))dx ,
b
VeS (a, b) = Ei EIa,b (X) [max {(1 + i) (u(a, b) + P (a, b)) Ia,b (X); 0} | i ]
Z1
= Ei (1 y) dFIa,b (y) ,
0
where
within model M1
VT (u, a, b) = Ei EIa,b (X) [max {i (u + P (a, b)) + P (a, b)
Z2
Ia,b (X); 0} | i ]] = Ei (2 y) dFIa,b (y) ,
0
Chapter 3 91
where
2 = iu + (1 + i)P (a, b) = 1 u
Z
= i VaR [X] + u1 (1 + (b, ))a (1 F (x))dx
b
Z
+(1 + )E[X] (1 + (b, ))a (1 F (x))dx,
b
where
within model M1
From here we obtain the total return on risk capital u is equal to:
Chapter 3 92
within model M1
V (u, a, b)
1 + (u, a, b) = = Ei [u,a,b (i)] (3.3.3)
u
R1 R2
(1 y) dFIa,b (y) (2 y) dFIa,b (y)
= Ei 0 0
,
u
within model M2
Ve (a, b)
1 + (a, b) = = Ei [a,b (i)] (3.3.4)
u(a, b)
R1
R2
(1 y) dFIa,b (y) (2 y) dFIa,b (y)
= Ei 0 0
u(a, b)
To avoid the degenerate situation in the revised model M2, where purchase
of full reinsurance offsets the required risk capital u(a, b) and underwriting
liability to zero (insurer assumes no insurance risk), we restrict quota share
in the class of change-loss reinsurance by an upper bound a1 < 1. We
investigate the value of the return (a, b) on risk capital in model M2 in
the following three regions {a [0, a1 ]} {b VaR [X]}, {a [0, a1 ]}
{VaR [X] < b (1 + i)VaR [X]} and {a [0, a1 ]} {b > (1 + i)VaR [X]},
since
(1 + i) (ab + (1 a)VaR [X]) (> b), if b VaR [X];
1 = (1 + i)VaR [X] (> b), if VaR [X] < b (1 + i)VaR [X];
(1 + i)VaR [X] (< b); if b > (1 + i)VaR [X].
Chapter 3 93
For a [0, 1)
Z1
(1 y) dFIa,b (y) =
0
1 ab
Rb R
1a
F (y)dy + (1 a) F (y)dy, if b VaR [X];
0 b
1 ab
= Rb R
1a
F (y)dy + (1 a) F (y)dy, if VaR [X] < b (1 + i)VaR [X];
0 b
R1
F (y)dy, if b > (1 + i)VaR [X]
0
Rb
F (y)dy + g1 (a, b; , i), if b VaR [X];
0
Rb
= F (y)dy + g2 (a, b; , i), if VaR [X] < b (1 + i)VaR [X];
0
R [X]
(1+i)VaR
F (y)dy, if b > (1 + i)VaR [X];
0
where
iab+(1+i)(1a)VaR [X]
Z
1a
g1 (a, b; , i) = (1 a) F (y)dy;
b
(1+i)VaR [X]ab
Z
1a
g2 (a, b; , i) = (1 a) F (y)dy,
b
and
Z2
(2 y) dFIa,b (y)
0
2 ab
Z2 Zb Z1a
= 1{2 b} F (y)dy + 1{2 >b}
F (y)dy + (1 a) F (y)dy
.
0 0 b
Chapter 3 94
Z1 Z1
lim (1 y) dFIa,b (y) = (1 y) dFI1,b (y)
a1
0 0
Rb
F (y)dy + (1 b) , if b VaR [X];
0
Rb
= F (y)dy + (1 b) , if VaR [X] < b (1 + i)VaR [X];
0
R1
F (y)dy, if b > (1 + i)VaR [X].
0
1
If we let y = 1a
, then for b (1 + i)VaR [X], using LHopitals rule, we
obtain
1 ab R
(y1 (y1)b)
Z
1a F (y)dy
b
lim (1 a) F (y)dy = lim
a1 y+ y
b
= lim (1 b)F (y(1 b) + b) = 1 b.
y+
Z2 Z2
lim (2 y) dFIa,b (y) = (2 y) dFI1,b (y).
a1
0 0
To proceed further we will consider the model under the assumption that
the aggregate amount of insurance claims X is exponentially distributed,
that is F (x) = 1 ex , x 0, > 0. As it was shown in the preceding
section VaR [X] = ln(1)
. We will also use the same formulae for the
premium income P (a, b) and the required risk capital u(a, b) defined earlier
in (3.2.14) and (3.2.15) respectively.
Now, we can determine explicit forms of a,b (i) from (3.3.4) on the fol-
lowing two ranges:
Chapter 3 95
1) b D {b (1 + i)VaR [X]}
1 b
1 a b 1 ab
1a
a,b (i) = 1 1e e e
1 b
1 a b 2 ab
1a
2 1e e e 1{2 b}
1 2
1
+ 2 1e 1{2 <b} ,
u(a, b)
where 1{A} is an indicator function of event A,
(1 + i) ab (1 a) ln(1) , if b VaR [X] = ln(1) ;
1 =
(1 + i) ln(1)
, if VaR [X] < b (1 + i)VaR [X];
where
ln(1 ) 1
g3 (a, b; , i) = i ab (1 a) + (1 + ) a(1 + (b, ))eb ;
ln(1 ) 1
g4 (a, b; , i) = i + (1 + ) a(1 + (b, ))eb ,
1.2
1 +
1.1 300
0 200
b
0.2
0.4 100
0.6
a
0.8
0
Figure 3.6: Graphical illustration of the total return 1 + (umin , a, b) in model M1 with
corporate tax = 30%
where
ln(1 ) 1 + 1+
b
1 = (1 + i) + u1 ae ,
ln(1 ) 1 + 1+
b 1+ b
2 = i + u1 ae + 1 ae 1+
and 2 = 1 u.
We consider numerical examples using the same parameters as in the
previous subsection, i.e. = 0.975; = 0.4, a1 = 0.92 (upper bound of
quota share in the class of admissible change-loss reinsurance contracts) and
1
= 0.01 (E[X] =
= 100). We assume that i is a deterministic value and
is equal to i = 10%. Let us further consider a corporate tax rate = 30%
for model M2. For model M1 we will consider the range of corporate tax
from 15% to 40%.
Figure 3.6 represents the graph of total return 1 + (umin , a, b) (u1 = 0)
on risk capital in the revised model M1 under = 30%. In this graph we
can see that there is demand for stop-loss reinsurance.
Chapter 3 97
1.22
1.25
1.215
1.2
1.21
1.15 1.205
b
b 600 800 1000 1200 1400
50 100 150 200 250 300 350
1.195
On the left hand side we have the graph that shows demand for stop-
loss reinsurance: the optimal retention of stop-loss reinsurance is equal to
b = 93.73 and the corresponding local maximum of return on equity is equal
to (umin , 1, b ) = 26.01%. On the right we have the graph 1 + (umin , a, 0)
on the interval b [VaR [X], ) that indicates that there is no demand for
reinsurance and local maximum of return on equity on this interval is equal
to 22.03%. Therefore, the global maximum of return on equity in model M1
with corporate tax 30% is equal to 26.01%.
The following table7 shows optimal reinsurance strategies in the model
M1 under different levels of corporate tax.
Table 3
Optimal reinsurance Maximal return on equity
15% b = or a = 0 26.83%
20% b = or a = 0 25.492%
25% b = 99.31 and a = 1 26.47%
30% b = 93.73 and a = 1 26.01%
35% b = 87.69 and a = 1 25.302%
40% b = 82.07 and a = 1 24.58%
1.25
400
1 + 1.2
1.15
300
1.1
1.05
0 200
b
0.25
0.5 100
a 0.75
10
Figure 3.7: Graphical illustration of the total return 1 + (umin , a, b) in model M1 with
corporate tax = 15%
1.3
1.215
1.28
1.21
1.26
1.24 1.205
1.22
b
600 800 1000 1200 1400
b
50 100 150 200 250 300 350
1.195
Chapter 3 99
1.3
1 + 1.25
1.2 300
1.15
0 200
b
0.2
0.4 100
0.6
a
0.8
0
Figure 3.8: Graphical illustration of the total return 1 + (a, b) in model M2 with
corporate tax = 30%
On the left hand side we can see that there is demand for change-loss
reinsurance, i.e. b = 58.41, a = 0.92 and corresponding local maximum for
return on equity is equal to (a , b ) = 31.02%. On the right we see that it
is optimal not to buy reinsurance for b [VaR [X], ) and local maximum
of return on equity on this interval is equal to 22.03%. Therefore, the global
maximum of return on equity in model M2 with corporate tax 30% is equal
to 31.02%. Note that the demand for reinsurance in the model M2 with
corporate tax is higher than demand in the same model without corporate
tax.
Comparing the two models M1 and M2 with corporate tax rate = 30%
we conclude that both models induce demand on reinsurance, however the
maximum return on equity in model M2 is higher than the analogous value
in model M1. The latter can be explained in terms of insurer capitalization.
In model M1 an insurer is more capitalized, since this model does not allow
insurer to reduce risk capital after taking reinsurance. On the other hand
Chapter 3 100
model M2 does allow insurer to reduce risk capital after taking reinsurance
and thus insurer is less capitalized. Holding extra risk capital reduces the
maximum value of return on equity.
Varenne (1997 [23]) and Jarrow and Purnanandam (2004 [88]), and refer-
ences therein). Insurance (underwriting) risk increases the probability that
an insurer will experience financial distress. Financial distress can be costly
due to both direct costs, such as legal fees (third party costs), lost value
from distressed sales (fire sale losses), and indirect costs, mainly loss of
reputation and franchise value. Some empirical studies (e.g. Opler and Tit-
man (1994 [113]), Andrade and Kaplan (1998 [2])) of financial companies
have revealed that financial distress results in costs of around 10 20% of
market value of assets. These costs are likely to be higher in the insurance
industry due to the credit-sensitive nature of policyholders.
Insurers with a substantial value of insurance outstanding claim liabilities
will be exposed to a higher risk of being financially distressed. Therefore,
highly leveraged insurers have an incentive to control their risk level. This
can be achieved by either holding a greater amount of risk capital (i.e. excess
of minimum amount required by regulator) or by the use of risk transfer, such
as traditional reinsurance or alternative risk transfer (ART) products (Culp
(2002 [34])). However, holding an extra amount of capital in an insurance
company is costly due to other frictions and market imperfections, such as
agency costs, costs arising from adverse selection and moral hazard, and
regulatory costs (see Merton and Perold (1998 [103])). Because of these
and other capital costs, insurers may avoid holding more than the minimal
amount of risk capital required by a regulator to reduce the probability of
financial distress event. Hence, risk transfer can be preferable. An additional
benefit of efficiently controlling risk via risk transfer (reinsurance) is that it
reduces the level of unnecessary volatility in the profit and loss statement.
Reducing volatility helps insurers to create sustainable shareholders value.
This section continues the research on risk management incentives in a
single period model with frictional costs, started in Krvavych and Sherris
Chapter 3 102
(2004 [94]), and considers risk management in two different single period
models of the insurance underwriting process with financial distress (FD)
costs. In the first model we consider the insurers solvency ratio, and in the
second model we consider risk based capital. In the first model we model the
dynamics of the companys liabilities and its solvency ratio with geometric
Brownian motions. Assuming that all liabilities are paid at the end of the
period, the insurer becomes financially distressed at the end of the period if
the solvency ratio falls below a pre-specified FD barrier. Insolvency occurs if
the solvency ratio falls below one. The objective function in this model is to
maximize the expected shareholders terminal payoff (valuable shareholders
claim on the terminal value of companys assets net of liabilities and FD
costs).
In the second model we model the companys surplus (net-worth) with a
geometric Brownian motion, and consider the underwriting process over the
period of time between two consecutive audits. At the beginning of the pe-
riod the company is solvent, i.e. it satisfies minimum solvency requirements
(minimal risk capital capacity or regulatory surplus) imposed by a regulator.
The firm is financially distressed if the net-worth of the company falls below
a threshold (barrier of financial distress, usually lower than required minimal
level of risk capital) during the period of time between audits. Insolvency8
occurs at the end of the period (on an auditing date) if the terminal net-
worth of the company is below the required minimal level of risk capital. We
consider the terminal value of excess of the required regulatory surplus, and
assume that a solvent company retains a part of this excess of regulatory
surplus (i.e. retained ratio times excess of surplus), and the rest is paid to
shareholders as dividends. We maximize the expected terminal value of the
8
We use term insolvency to define the economic state in which an insurance company
is undercapitalized from a regulators point of view and cannot continue its underwriting
process without reorganizing its capital structure.
Chapter 3 103
3) constant;
In the case of FD of the fourth form we assume that in the event of financial
distress, the companys operations are adversely affected in such way that
it is unable to realize the full upside potential of its surplus. Such repre-
sentation of FD costs is consistent with the idea that a financial distressed
insurance company is unable to capitalize on its real options and suffers
reduced growth as a result. Although this form of FD costs is rather theo-
retical (technical) than practical, the use of it has advantages in obtaining
an analytical (exact) form of the optimal value of the companys aggregate
investment-underwriting risk as a function of the financial distress barrier,
the deadweight losses caused by financial distress, the minimal value of sur-
plus required by a regulator and the length of the period. At the same time
we can detect incentives to control the companys aggregate investment-
underwriting risk in the models of the maximization of shareholders value
under the presence of FD costs of other three forms using numerical calcu-
lations only.
In the first model we consider the deadweight losses caused by financial
distress that are proportionate to the terminal value of the companys assets.
Chapter 3 104
In this subsection we define two solvency models: the solvency ratio (SR)
model and the risk based capital (RBC) model.
The SR model
Similarly to Sherris and van der Hoek (2004 [136]) we consider an insurers
underwriting process in the period of time from 0 to T and model the risk-
neutral dynamics of insurers liabilities and solvency ratio with geometric
Brownian motions.
Denote the value of companys liabilities at time t by Lt for t [0, T ].
Assume that the Q-risk-neutral dynamics of Lt are
and that there are no claim payments made other than at the end of the
period, i.e. LT . If in addition L can be replicated by traded assets then
L = r, the risk free rate.
We know that the insurance policies are contingent claims on the value of
the liabilities with payoff that depends on the insurer solvency and assume
that in the SR model the solvency is measured by solvency ratio
At
t = ,
Lt
where At denotes the value of the companys assets at time t [0, T ]. We
assume that the Q-risk-neutral dynamics of t are
with
A0
0 = > 1,
L0
i.e. the insurer is solvent at the beginning of the period. As in Sherris and
van der Hoek (2004 [136]) it is assumed that the parameters and are
known and given (estimated from data).
We define three different economic states of the insurance company:
financially healthy;
insolvent.
The insurer becomes financially distressed if the terminal value of the sol-
vency ratio T falls below a pre-specified threshold (FD barrier) b (i.e.
T b). In the state of FD the company experiences deadweight losses
proportionate to the terminal value of assets AT with proportionate coeffi-
cient 1 w, w [0, 1). Being financially distressed the insurer is solvent if
the terminal value of assets net of FD costs exceeds the terminal value of
1 1
liabilities, i.e. when w AT > LT or T > w
. If w b < 1 (or b < w
) then
financial distress implies immediate insolvency, and thus the two states fi-
nancial distress and insolvency coincide. Another consideration is that the
insurer holds risk capital as a part of its total assets to back up its liabilities
and hence the solvency ratio of financially healthy insurer is considerably
higher than 1. If the solvency ratio is quite close to 1, then more likely the
insurer experiences shortage in the necessary risk capital, i.e. it is financially
1
distressed. Therefore, we assume that b > w
which also ensures that there
is a chance of recovering from the FD state. It is obvious that if the insurer
1
is financially healthy (i.e. T > b) then it is solvent, since T > b > w
> 1.
Chapter 3 106
V0SR = EQ erT (w AT LT )+ 1{T b} + (AT LT ) 1{T > b}
" ( + )#
1
= EQ erT LT w T 1{T b} + (T 1) 1{T > b}
w
" ( +
rT 1
= EQ e LT w T
w
(w T 1) 1{T > b} + (T 1) 1{T > b}
" ( + )#
1
= EQ erT LT w T + (1 w) T 1{T > b} . (3.4.3)
w
The main objective function within this model is to maximize V0SR with
respect to the insurers risk = (L , ). The insurer can adjust the value
of risk to the optimal value = arg max V0SR by:
immediately drop to zero (the company is liquidated), and when the insurer
is solvent on the maturity date, it retains the excess of minimal level of
surplus (regulatory capital) at the constant retention (plow-back) rate and
the rest is paid to shareholders as dividends. So the maximization of the
terminal value of dividends is equivalent to the maximization of the expected
terminal value of the companys surplus net of regulatory surplus.
In contrast to the SR model, where we assumed that all the insurers
liabilities are paid at the end of the period, the liabilities within the RBC
model are assumed to be paid continuously over the period. We define the
insurers economic state as financially distressed in the following way. If the
insurers value of its surplus S never hits a pre-specified financial distress
barrier, say D, over the period [0, T ], the terminal surplus value is ST . If the
financial distress barrier is hit, the insurer incurs deadweight losses and the
terminal surplus value falls to F (ST ) < ST . Here, the function F represents
the terminal surplus value net of FD costs. Let mST denote the minimum
value of surplus over [0, T ] (i.e. mST = min St ). We assume that the insurer
t[0,T ]
experiences financial distress when its surplus hits a pre-specified level which
is lower than regulatory surplus, i.e. D < S .
If the insurance company was not financially distressed (mST > D) and
it is solvent on the terminal date (ST > S ), then the terminal value of the
excess of the minimal level of the companys surplus is ST S . The terminal
value of the excess of minimal level of surplus is equal to F (ST ) S > 0 in
the case when the company was financially distressed but it remains solvent
on the maturity date.
From here we can calculate the expected present value of the companys
Chapter 3 108
We can see in (3.4.5) that the value V0 has three components. The first term
EQ [(ST S )] represents the unconstrained surplus value of the company net
h i
of regulatory capital. The second term EQ (ST F (ST )) 1{mST D}{F (ST )>S }
represents the deadweight losses caused by financial distress. The sharehold-
ers of a financially distressed but solvent insurance company bear financial
distress costs and thus the terminal value of surplus net of regulatory capital
is reduced by this amount. These costs can be reduced by risk management
strategies. Existence of the third positive term in (3.4.5) can be increased
by risk-taking strategies by shareholders. By increasing the companys inte-
grated investment-underwriting risk, the shareholders can make themselves
better off by increasing the call option value. But at the same time the
expected losses in the event of financial distress also increases with the risk.
Therefore, the optimal level of integrated investment-underwriting risk is
determined by the trade-off between these two incentives.
Chapter 3 109
= (L (1 P ) (L + P )) Lt dt + sL dWtL ,
where L > 0 and P > 0 denote, respectively, the expense ratio for admin-
istrative expenses that are proportional to expected losses and the expense
ratio for acquisition expenses that are proportional to premiums; the volatil-
ity of this diffusion is assumed to be sL = L Lt .
By definition the insurers assets are equal to the insurance liabilities (loss
reserves) plus surplus, and thus investment income is generated by investing
Chapter 3 111
these assets in the capital market. It is assumed that the total investment
return follows a geometric Brownian motion with drift rI and volatility I ,
i.e. the instantaneous investment income inflow at time t is
Lt
Denote by t = St
the current leverage ratio at time t. For the sake of
simplicity and ability to illustrate results we assume that W I and W L are
independent, the insurer can control its leverage ratio over the period to
keep it constant at level . Now, taking into account the fact that the
instantaneous insurers surplus at time t, dSt , is the sum of instantaneous
values of investment income and underwriting profit (i.e. d(It + t )) we can
write the diffusion process of the insurers surplus as
dSt = d(It + t ) = (L (1 P ) (L + P )) Lt dt
= St dt + St dWt , (3.4.6)
p
where = (L (1 P ) (L + P ) + rI ) + rI , = 2 L2 + ( + 1)2 I2 ,
and W is a standard Brownian motion associated with the surplus.
So, we come up with a model of the insurers surplus described by a
geometric Brownian motion under the physical probability measure P. This
means that within the RBC model the companys surplus is always positive.
This is a difference between this and the SR model, where the surplus,
defined as a difference between the values of assets and liabilities, can take
negative values in the case of insolvency. However, it should be noticed that
the use of the log-normal model of the surplus can be meaningful within
the RBC model, since the insurers insolvency occurs at a positive level of
capitalization, or in other words, it is defined by the event where the insurers
surplus hits a positive threshold, i.e. value of the regulatory surplus.
Chapter 3 112
In the case of taking quota share proportional reinsurance with the cedents
retention level (0, 1) the drift and diffusion of S in (4.4.3) are respectively
equal
= [L (1 P ) (L + P ) + rI pre (1 )] + rI and
p
= 2 2 L2 + ( + 1)2 I2 , where pre is the reinsurance premium rate.
In order to calculate the value V0SR in (3.4.3) we change the numeraire from
the risk free bank account to Lt and correspondingly change the measure
from the risk-neutral measure Q, defined under the old numeraire, to the
e Under the risk-neutral measure Q
new one Q.
LT = L0 eL T ZT ,
where
L 1 2
ZT = eL WT 2 L T is a martingale with respect to Q.
e is defined as follows
The new risk-neutral measure Q
dQe
FT = ZT and t [0, T ] EQ [ZT | Ft ] = Zt .
dQ
Then for any function of the terminal value of T using Girsanovs theorem
we obtain
r(T t) ZT
EQe e r(T t)
f (T ) Ft = EQ e f (T ) Ft
Zt
and thus
EQ er(T t) LT f (T ) Ft = EQe er(T t) f (T ) Ft Zt L0 eL T
= Lt eL (T t) EQe er(T t) f (T ) Ft .
f = W L L
W t t
Chapter 3 114
e
is a Q-Brownian motion, where L is the instantaneous correlation between
the value of liabilities and solvency ratio, defined by
e
Therefore, the new Q-risk-neutral dynamics of are
dt = t dt + t dWt
= t dt + t dWft + L dt
=
e t dt + ft
e t dW
= (r (r
e )) t dt + ft ,
e t dW (3.4.7)
where
e = + L L and
e = .
Now, the expected present value V0SR of shareholders terminal payoff in
(3.4.3) can be rewritten in the following way
" ( + )#
SR rT 1
V0 = EQ e LT w T + (1 w) T 1{T > b}
w
( " + #
L T rT 1
= L0 e w EQe e T (3.4.8)
w
o
+(1 w)EQe erT T 1{T > b} .
e
According to the Q-risk-neutral dynamics of in (3.4.7) the value
c0 = EQe erT (T K)+
SR
V0 = L0 0 ee T ( (h+ ) (l+ )) > 0, since h+ > l+ .
w
This means that the value V0SR increases with a decrease in the FD costs (or
with an increase in w). It is easy to see that the third term in (3.4.10) de-
creases when the FD barrier b increases. Therefore, the value V0SR decreases
with an increase in the FD barrier b.
We also consider the problem of finding an optimal volatility of the
solvency ratio under which the value V0SR takes its maximum. To do so we
Chapter 3 116
F2 (ST ) = ST C,
F3 (ST ) = ST (ST (S + U ))+ ,
where U > 0 is the parameter of FD costs. The higher the value of U , the
higher value of upside potential of ST , or, equivalently, the lower value of
FD costs.
We rewrite the value V0RBC from (3.4.5) for all the three forms of FD
costs. In the first case where F = F1
(1)
V0RBC = V0 (3.4.11)
" #
+
S
= erT EQ (ST S )+ 1{mST >D} + w ST 1{mST D} ,
w
(2)
V0RBC = V0
h i
rT + +
= e EQ (ST S ) 1{mT >D} + (ST (C + S )) 1{mT D}
S S
h
rT
= e EQ (ST S )+ (ST S )+ 1{mST D} (3.4.12)
i
+ (ST (C + S ))+ 1{mST D} ,
As we can see in all three cases the value V0RBC is represented by the values
of two different barrier options. In the third case there is a term representing
the joint distribution of the minima and the terminal value of the companys
surplus that follows a geometric Brownian motion.
Chapter 3 118
To evaluate V0RBC we will use the well known approach for the pricing of
path-dependent options or barrier options (e.g. see Etheridge (2002 [54]) or
Musiela and Rutkowski (1998 [108])). We know that under the equivalent
martingale measure Q, St = S0 eYt , where
r 12 2
Yt = t + WtQ
and WtQ is a Q-Brownian motion. It is also well known from financial calculus
that for Yt = bt + WtQ , and mt = min Ys
s[0,t]
(
pT (bT, x)dx, x < a,
Q [mT a, YT dx] =
e2ab pT (2a + bT, x)dx, x a,
(xy)2
where pt (x, y) = 1 e 2t is the Brownian transition density function
2t
h i
(1) rT rT + +
V0 = EQ e (ST S ) EQ e (ST S ) 1{mST D}
" #
+
S
+w EQ erT ST 1{mS D} ,
w T
and in (3.4.12) it is
h i
(2) rT rT + +
V0 = EQ e (ST S ) EQ e (ST S ) 1{mT D}
S
h i
rT +
+ EQ e (ST (C + S )) 1{mST D}
i.e. in the first two cases it is equal to the value of a call option plus the
difference of two down-and-in-call options with different strike prices.
Chapter 3 119
1 D
For any strike price K > 0 and a =
ln S0
the value of a down-and-
in-call option is
h i
rT
EQ e (ST K) 1{mST D} = EQ erT (ST K)+ 1nm
+
1
o
ln SD
T
0
Z
= erT (S0 ex K) Q [mT a, YT dx]
1 K
ln S0
Z
= erT (S0 ex K) e2ab pT (2a + bT, x) dx
1 K
ln S0
1 K
We have used the fact that D < K, and thus x x0 =
ln S0
>
1 D
ln S0
= a.
We calculate the first integral
Z
erT S0 ex e2ab pT (2a + bT, x) dx
x0
Z
rT 2ab 1 (x (2a + bT )2 2xT )
=e S0 e exp dx
2T 2T
x0
Z
rT 2 T
+2a+bT +2ab 1 y2
=e S0 e 2 e 2 dy
2
x0 (2a+bT )T
T
2r2 1 D2 2
D D2 ln KS0
+ r+ 2
T
,
=
S0 S0 T
rT 2ab 1 y2
= Ke e e 2 dy
2
2
2r2 1 ln KSD
+ r 2
T
D
0 2
= K erT
S0 T
and
2r2 1 2 2
(2) D D D
V0 = C (S0 , 0, S , T ) C , 0, S , T C , 0, C + S , T ,
S0 S0 S0
Using the same approach of barrier options valuation we can calculate the
(3)
value V0 . To find an optimal
b under which the value function V achieves
its maximum one has to solve FOC. In general case where r 6= 0 it is difficult
to obtain explicit form of optimal and the numerical methods must be used.
Chapter 3 121
To find an extremum
b of the function V0 () first we have to calculate
Chapter 3 122
(3)
The corresponding score equation V
0
= 0 can be rearranged to the linear
one with respect to 2 :
(3)
V =0
0
2 2
S0 2 2 D2 2
ln + T 2 T ln S0 = ln
+ T 2 2 T ln D
S 2 S0 (S + U ) 2
D S 2 D 2
2 1 ln S02 (S +U ) ln S (S +U )
b =
T ln S S+U
2 (3)
It can be directly verified that the second-order derivative V
2 0
is negative
at =
b. So, we can state that
(3)
b = arg max V0 ()
As we can see the optimal insurers risk depends on the FD barrier D, the
initial value S0 of companys surplus, the regulatory capital S , the param-
eter U of FD costs and the length T of the period between two consecutive
Chapter 3 123
= S +U
T (S + U ) ln2 S
S +U
1 ln2 S 4 ln SD0 ln SD
= S +U
. (3.4.15)
T (S + U ) ln2 S
q
S0 S
2 0 2 ln ln
From (3.4.15) we can see that U
b > 0 iff U > U = e D D 1 S >
0. That is, when FD costs are extremely high (or the FD parameter U is
close to the level of insolvency S ) then the states insolvent and finan-
cially distressed are almost the same and this is the situation where it is
Chapter 3 124
too late to control risk (i.e. the company is nearly under liquidation). How-
ever, for reasonable values of FD costs, i.e. for all U > U 0 > S , we showed
that the optimal companys risk decreases with increase of FD costs (or with
decrease of the FD parameter U ).
We end this subsection by considering the question as to how insurance
risk managers can adjust the value of risk to the optimal level
b. From
our model of the companys surplus we can see that the integrated risk is
q
= 2 2 L2 + ( + 1)2 I2 .
3.5 Conclusion
In this chapter, we have investigated the risk management incentives in in-
surance companies in the presence of such frictional costs as corporate tax
and financial distress costs. We started from an analysis of a demand for
change-loss reinsurance in two single-period models of shareholders value
creation. In both models the gross insurer premium is determined using an
expected value premium principle that does not reflect the effect of insol-
vency on policy payoff. In the first model the required minimal risk capital
is predetermined at the beginning of the period without taking into account
possible purchase of reinsurance. In the second model an insurer is allowed
Chapter 3 125
to reduce its risk capital to the level under which the minimum solvency re-
quirements are satisfied. We showed that there is no demand for reinsurance
in the first, more conservative model without frictional costs. However, un-
der the presence of frictional costs, such as corporate tax, this model induces
demand for reinsurance.
At the same time it was shown that the second model induces demand
for reinsurance. In the frictionless environment this model has an optimal
trade-off between the required minimal level of the risk capital and purchase
of reinsurance. There is also demand for reinsurance in the second model
under the presence of corporate tax.
The demand for reinsurance in the second model under the absence of
frictional costs is likely due to the assumption of an actuarial premium prin-
ciple, according to which the premium is not adjusted with respect to the
value of insolvency exchange option.
We have also analyzed the demand for reinsurance in models with fi-
nancial distress costs. We considered modelling of insurer solvency in the
presence of four different forms of financial distress costs, and in the case
of RBC solvency model with financial distress costs in the form of reduced
upside potential of the terminal value of insurers surplus we demonstrated
that there are risk management incentives to control underwriting and in-
vestment risks through reinsurance and investment hedge. We showed that
these insurance risk management incentives increase with an increase of fi-
nancial distress costs.
The results of this chapter are very important and show that the decision
to reinsure can be treated as both a risk-management and a capital-structure
tool in shareholders value creation.
Chapter 4
4.1 Introduction
The main goal of this chapter is to reconsider static single period models
of integrated risk management for an insurance company, discussed in the
previous chapter, in a dynamic setting. It is worth noticing that in classic
126
Chapter 4 127
setting for risk reserve process. This can be explained by the impossibility
of finding an explicit solution to this kind of problem. Indeed, Dassios
and Embrechts (1989 [39]), for instance, determine the optimal dividend-
payment policy just for the class of all policies with a barrier, and Schal
(1998 [125]) proves the existence of an optimal feedback control and verifies
that it is a maximizer of a specific Hamilton-Jacobi-Bellman equation.
The application of controlled diffusion models to solving the same divi-
dend optimization problem gave more fruitful results. The pioneering and,
perhaps, most fundamental paper in this research area was the paper by
Shreve et al (1984 [137]), where the authors considered a stochastic con-
trol absorption problem for which a general diffusion process of the storage
is controlled by subtracting a nondecreasing withdrawal process. The con-
trolled process is absorbed if it reaches zero, and the objective function to be
maximized is the expected discounted value of the withdrawals. Obviously,
this general approach is readily applicable to the dividend maximization
problem in insurance. Here the absorption barrier stands for the ruin of an
insurer, and withdrawals stand for dividend-payments. The authors showed
that the optimal withdrawal policy is a barrier policy. Continuing this work
Martin-Lof (1994 [101]), Brockett and Xia (1995 [24]), Jeanblanc-Picque
and Shiryaev (1995 [89]) and Asmussen and Taksar (1997 [8]) considered
maximizing the expected value of discounted dividend-payments paid until
time of ruin for a liquid risk reserve of the insurance company modelled by a
Brownian motion with constant drift and diffusion coefficients. They showed
that the optimal dividend-payment policy is a barrier policy.
When the liquid risk process of the insurance company is modelled by
a Brownian motion with drift, the drift term corresponds to the expected
profit per unit time, while the diffusion term is interpreted as risk. The larger
the diffusion coefficient the greater the insurance risk the company carries.
Chapter 4 130
The company can decrease the underwriting risk through purchasing rein-
surance, i.e. it can dynamically control risk (with reinsurance) to decrease
the diffusion coefficient. This sets the new dividend optimization-risk con-
trol models with two control variables: dividend-payments and reinsurance
policies. One may raise the question whether there is an incentive to buy
reinsurance, since doing so an insurer decreases simultaneously the drift
coefficient (profit potential) as well. In the following research papers by
Hjgaard and Taksar (1999 [82]; 2001 [83]), Asmussen et al (2000 [9]) and
Choulli et al (2003 [28]) the authors showed that there may be a demand for
reinsurance in maximizing expected value of discounted dividend-payments
paid until time of ruin. This can be explained intuitively in the following
way. Taking reinsurance increases the expected value of ruin time (i.e. in-
creases companys lifetime), which may lead to increasing the companys
objective function.
Another very popular risk optimization criterion for insurance compa-
nies, in particular among European researchers, is to control the risk process
by reinsurance so that the ruin probability of ruin is minimized. In recent
years many papers have been written on this topic, and among them are
Schmidli (2001 [128]; 2002 [129]), Hipp and Taksar (2000 [80]), Hipp and
Plum (2000 [77]; 2003 [78]), Hipp and Vogt (2003 [81]), Hipp and Schmidli
(2003 [79]), Taksar and Markussen (2003 [144]), and Hipp (2003 [76]). As it
was already explained on page 127, minimizing the ruin probability is not
the sole insurer objective. Indeed the insurer strives to maximize sharehold-
ers value. However, at the same time the level of underwriting risk, which
is traditionally measured by ruin probability, is also important, since the
insurer operates under solvency constraints imposed by a regulator. There-
fore, the basis of insurer risk management is to find a trade-off between risk
and profitability.
Chapter 4 131
under the solvency restrictions in Section 4.4, and in the presence of insol-
vency costs in Section 4.5.
Let us start with the mathematical formulation of the Levy model. As in the
classical actuarial risk model of Cramer and Lundberg we suppose that in a
dynamic setting the insurance company receives a deterministic amount of
premium income at the rate c > 0 and claims arrive according to a Poisson
process Nt with intensity > 0. The size of the kth claim is modelled by
the random variable Yk 0. We suppose that the {Yk , k 1} are i.i.d. with
Chapter 4 133
R
distribution function F and that the mean m = y dF (y) and the variance
0
R
s2 = (y m)2 dF (y) are finite. All random elements are defined on a
0
common probability space (, F, P) with filtration (Ft )t0 that is induced
by Nt and Y1 , ..., YNT . The aggregate claim over the period [0, t] is modelled
P
Nt
by a compound Poisson St = Yi . It is assumed that all of the surplus
i=1
is invested in a stock market instrument, whose price is governed by the
geometric Brownian motion
dIt = rI It dt + I It dWtI
dR0 = x, (4.2.2)
Chapter 4 135
P
Ns
where dSt (a (t), b (t)) = dSs (a (t), b (t))|s=t = d I1a (t), b (t) (Yi )|s=t .
i=1
We define the corresponding companys ruin time by
= inf {t 0 : Rt 0} .
For all admissible control policies we define the return (value) function
as
Z
V (x) = E et lt dt, (4.2.3)
0
where is the discounting rate (usually hurdle rate). The optimal value
function is defined as
That is, the companys objective is to find an optimal policy such that
V (x) = V (x) for all x. One can recognize this as a stochastic optimal
control problem and try to solve it using Hamilton-Jacobi-Bellman (HJB)
equation (see Fleming and Rishel (1975 [56]) and/or Fleming and Soner
(1993 [57])). Unfortunately, it is not easy to solve in the general Levy
setting (4.2.1), and so far no attempts to do this have been published in
the research literature. The resulting HJB equation becomes an integro-
differential equation, whose closed form solution would be difficult to obtain
if at all possible.
In the case when I = 0 the model of the companys risk reserve is reduced
to the piecewise deterministic Markov process (PDMP). There have been a
few attempts to find the solution in a PDMP framework. It turns out that
within this framework the control is difficult, and an explicit solution cannot
Chapter 4 136
be found. Therefore, the first researchers, Dassios and Embrechts (1989 [39])
who tried to find the solution, considered uncontrolled dividend policy and
determined the optimal dividend policy among all barrier policies. Later
Davis (1993 [40]) developed the theory of existence of the solution and Schal
(1998 [125]) proves the existence of an optimal feedback control only to the
considered problem with no reinsurance and debt liability.
We consider the case when the dynamics of the surplus process can be ap-
proximated by a diffusion. In general the diffusion approximation can be
suitable to apply at least in the case of a large portfolio. Motivations for a
diffusion approximation of the risk reserve can be found in Iglehart (1969
[85]), Emanuel et al. (1975 [53]), Grandell (1977 [72], 1978 [73], 1990 [74]),
Schmidli (1993 [127]).
We start the formulation of diffusion approximation from the following
probabilistic result:
n o
d
Rt t + WtS (4.2.5)
2 t0
in the space of right continuous functions with the left limits endowed with
Skorohod topology as 0, where WtS is a standard Brownian motion that
corresponds to the approximation of the dynamics of Cramer-Lundberg risk
process, and
= (a , b ) = lim 2
EI1a, b (Y ) + 2 (m k(b, )) [m EI1a, b (Y )]
0
2 EI1a, b (Y ) = EI1a, b (Y )
Z
= m (1 a) (1 F (y)) dy;
b
Chapter 4 137
2
2 = 2 (a , b ) = E I1a, b (Y )
Z Z Z
= 2 yF (y)dy (1 a2 ) yF (y)dy + (1 a)ab F (y)dy
0 b b
dR0 = x, (4.2.7)
In this particular case the dynamics of the companys risk process are given
by
dR0 = x (4.2.9)
Proof. For different initial values of the risk reserve x and y consider the
corresponding admissible controls x and y . Let 0 < < 1 and define the
control z by
and
Proposition 4.2.2. Let the components of any admissible control are de-
fined by 0 a 1 and 0 l B. Then the optimal value function V
satisfies the following HJB equation
1 2 2 00 0
max a V (x) + ( a l)V (x) V (x) + l = 0 (4.2.11)
a[0,1]; l[0,B] 2
V (0) = 0 (4.2.12)
Proof. At the first glance the function V satisfies the main dynamic pro-
gramming principle: t > 0
min{ , t}
Z
V (x) = sup E es l (s)ds + 1{t< } et V (Rt ) .
0
Chapter 4 140
Applying Itos Lemma (see, e.g., ksendal (2000 [112]) or Rolski et. al
(1999 [124])) to the second term (i.e. applying Itos Lemma to the function
f (t, x) = et V (x)) in the right hand side of the preceding inequality we get
Tu
h i Z b
u
E e b V RTbu = V (x) + E es A(a,l) V Rsb ds,
T
b
where
2 a2 00
A(a,l) [f (x)] = f (x) + ( a l)f 0 (x) f (x)
2
We substitute this result into the inequality, subtract term V (x) from both
sides and divide both sides by E [Tbu ] and get
u
T
Rb
E es l + A(a,l) V Rsb ds
0 u0
0 l + A(a,l) V R0b .
E [Tbu ]
Since this is true for arbitrary constants a and l we conclude that
0 max l + A(a,l) [V (x)] . (4.2.13)
a[0,1]; l[0,B]
The equation (4.2.11) follows from (4.2.13) and (4.2.14). Finally the bound-
ary condition V (0) = 0 follows from the fact that V (0) = 0.
To find a solution to HJB equation (4.2.11), we should use the fact that the
optimal value function V is concave due to Proposition 4.2.1, and that the
expression in the left hand side of HJB (4.2.11) is a linear function of l and
a. Therefore, for each x the maximizer of the left hand side of (4.2.11) with
respect to l is either l = 0 if V 0 (x) > 1 or l = B, if V 0 (x) 1. Since V is
concave, the set {x : V 0 (x) > 1} is an interval [0, x1 ). Using this we can
split the HJB (4.2.11) into two:
1) x < x1 :
1 2 2 00 0
max a V (x) + aV (x) V (x) = 0
a[0,1] 2
2) x x1 :
1 2 2 00 0
max a V (x) + ( a B)V (x) V (x) + B = 0
a[0,1] 2
0
For the first case we find maximizer a (x) = V (x)
2 V 00 (x) , which has to be a
fraction. At this stage we assume that there exists x0 such that x0 < x1 , and
Chapter 4 142
a (x) (0, 1) for all x [0, x0 ) and a (x) = 1 for all x x0 . Substituting
a (x) into HJB equation we obtain
[V 0 (x)]2
V 0 (x) = 0, x [0, x0 ).
2 2 V 00 (x)
B
V (x) = C4 e (B)x + .
Taking into account the assumption that the function V is the twice contin-
uously differentiable solution to the HJB equation, we obtain the following
boundary conditions
While deriving this solution we assumed that x0 < x1 . As we can see now
this assumption holds if and only if
( B) ()
ln > 0,
+ () ( B)
or equivalently, iff the maximal dividend rate exceeds some threshold, i.e.
2
B> + . (4.2.15)
2
In this case the optimal risk control policy is to retain the risk, the amount
of which is linearly proportional to the current amount of reserve, until the
risk reserve reaches the level x0 . That is the optimal cedents retention level
x
of quota share proportional reinsurance is equal to a (x) = x0
if x < x0 and
is equal to 1 if x x0 . At the same time the optimal dividend policy is
to pay dividends at the maximal rate when the risk reserve level exceeds x1
(x1 > x0 ).
In the case where (4.2.15) is not valid, that is x0 > x1 , if x0 < then
it can be shown that smooth fit conditions at this point fail. Therefore,
we conclude that x0 = and x1 is the only one switching point. On the
interval [0, x1 ] the form of the solution is similar to that one on (0, x0 ) in the
case where (4.2.15) is valid, that is V (x) = C 1 x . For x x1 the solution
Chapter 4 144
V must satisfy
1 2 2 00
(a ) V (x) + ( a B)V 0 (x) V (x) + B
2
V 0 (x)
2 00 = a < 1.
V (x)
The only pair (V, a ) that satisfies both preceding equations is
B
a = 2
2
+
and
B 2
V (x) = + C 2 e a x .
Using smooth fit conditions V (x1 ) = V (x1 +) and V 0 (x1 ) = V 0 (x1 +) = 1
we obtain x1 and the constants C 1 , C 2
B(1 ) x1 B B
x1 = ; C1 = 1 ; C2 = e x1 .
2
Therefore, in the case where B
+ 2
the solution to the HJB
equation (4.2.11) is equal to
x1 x , x < x1
V (x) = x1
B 1 e B
(xx1 )
, x x1 ;
the optimal risk control policy is to retain the risk, the amount of which is
x
linearly proportional to the current level of risk reserve, i.e. a (x) = x0
=
a xx1 , x < x1 , and retain the maximum amount a of risk when initial risk
reserve exceeds level x1 .
Finally we consider the case where the rate of dividends payout is un-
bounded. Using the arguments similar to those used in the proof of Proposi-
tion 4.2.2, and also elements of singular stochastic control theory (see Chap-
ter 8 in Fleming and Soner (1993 [57])) we can derive the corresponding HJB
equation, which the optimal value function V satisfies. This HJB equation
is given by
2 a2 00 0 0
0 = max max V (x) + a V (x) V (x) , 1 V (x) (4.2.16)
a[0,1] 2
Chapter 4 145
To solve the latter HJB equation we assume that x0 < x1 once again and
then solve the following reduced HJB equation for all x < x1 . Doing so, we
get the same form of the solution as we had in the previous case. For x > x1
the function V satisfies
V 0 (x) = 1.
we get C b1 = C1 , C b2 = C2 , C
b3 = C3 , C
b4 = x1 and x1 = x0 +
h i
1 ()
+ () ()
ln + () . Note that x1 is always greater than x0 , and thus
the initial assumption about x0 and x1 was correct. In this case the opti-
mal insurers retention level of the quota share proportional reinsurance is
n o
a (x) = max xx0 , 1 . That is the optimal risk control policy is to retain the
risk proportional to the current risk reserve x, until the risk reserve reaches
x0 , and retain the maximal risk when the risk reserve is above x0 . The op-
timal dividend policy is the barrier policy with barrier x1 , i.e. whenever the
risk reserve goes above x1 the excess is immediately paid out as dividends.
dR0 = x, (4.2.18)
Rb Rb
where (b) = F (y) dy, 2 (b) = yF (y) dy, and F is the cdf of the
0 0
individual claim size. The key idea for solving this particular problem is
to use parametrization of control variable b via the drift term . This can
be done in the following way. The function (b) is a continuous increasing
function of b on support [0, J] of cdf F , where J = sup x : F (x) > 0
. Therefore, there exists an inverse function b(). The function (v) =
2 (b(v)) is a strictly increasing convex function of v on interval [0, ], where
= m. So the control policy is now = (v , l ), where v [0, ].
The dynamics of the companys surplus is then
p
dRt = (v (t) l (t)) dt + (v (t)) dWt (4.2.19)
dR0 = x. (4.2.20)
Using the arguments similar to those used in the previous paragraph, we get
the HJB equation for the optimal value function V :
It turns out that the solutions to these two HJB equations depend on whether
the support [0, J] finite or infinite. Here we provide solutions for all possible
cases.
Chapter 4 147
(vB) (vB)2 +2 (v)
e
where (v) = (note b(v) is the inverse function of
(v)
e 1) = 1 ;
(v
b(v1 )
Rv (y) b0 (y)
and x1 = G(v1 ), where G(v) = 2 b2 (y)(y)+2yb(y)
dy.
0
In this case the optimal risk control is defined by the retention level
b (G1 (x)), if the current risk reserve level x is below x1 . If x x1 , then
retention level is constant and equal to b(v1 ). The optimal dividend policy
is to pay dividends at the maximum rate whenever x > x1 .
1.2) Bounded dividend rate and finite support.
a) If the upper bound of the dividend rates is enough high, namely if B >
2
J + 2
, then the optimal value function is
x
R0
Rx 1
dz
C ey
b(v(z))
dy, x < x0
0
V (x) =
C1 e ( )(xx0 ) + C2 e+ ( )(xx0 ) , x0 < x < x1 ;
B e ( B)(xx1 )
+ ( B)
, x x1 ;
Chapter 4 148
2 2 2
where x0 = G( ); ( ) = 2
;
+ ( ) + J1 ( ( ) ( B))
D= > 1;
( ) + J1 (+ ( ) ( B))
1
x1 = x0 + ln D > x0 ;
+ ( ) ( )
+ ( )
D + ( ) ( )
C1 = < 0;
( ) (+ ( ) ( ))
( )
D + ( ) ( ) ( ( ) ( B))
C2 = > 0;
+ ( ) (+ ( ) ( )) (+ ( ) ( B))
C = C1 ( ) + C2 + ( ).
In this subcase the optimal risk control is defined by the retention level
b (G1 (x)), if the current risk reserve level x is below x0 . If x x0 , then there
should be no reinsurance. The optimal dividend policy is to pay dividends
at the maximum rate whenever x > x1 .
b) If the upper bound of the dividend rates is not enough high, i.e. if
2
B J + 2
, then the optimal value function V coincides with the
one obtained in case 1.1).
1.3) Unbounded dividend rate and infinite support.
The optimal value function is
x
R1 1
Rx dz
ey
b(v(z))
dy, x < x1
0
V (x) = x
R1 1
Rx1 b(v(z)) dz
x x1 + e y dy, x x1 ;
0
where x1 = G( ).
In this case the optimal risk control is defined by the retention level
b (G1 (x)). Here, the optimal risk control policy requires to purchase some
excess-of-loss reinsurance. As in the the previous cases the optimal retention
level b (G1 (x)) is an increasing function, which means that with higher risk
reserve (i.e. when the company is more capitalized) we need less reinsurance.
The optimal dividend policy is the barrier policy with barrier x1 .
Chapter 4 149
0
V (x) =
C1 e+ ( )(xx1 ) + C2 e ( )(xx1 ) , x0 < x < x1 ;
x x1 + C 3 , x x1 ;
2
2 2
where x0 = G( ); ( ) = 2
;
1 ( ) (1 + J+ ( ))
x1 = x0 + ln > x0 ;
+ ( ) ( ) + ( ) (1 + J ( ))
(
C1 = ;
+ ( (+ ( ) ( ))
+ (
C2 = ;
+ ( (+ ( ) ( ))
C3 =
C = C1 + ( )e+ ( )(x1 x0 ) + C2 ( )e ( )(x1 x0 ) .
In this case the optimal risk control is defined by the retention level
b (G1 (x)), if the current risk reserve level x is below x0 . If x x0 , then
b(x) = J and there should be no reinsurance. The optimal dividend policy
is the barrier policy with barrier x1 .
dR0 = x. (4.2.22)
This problem was investigated by Hjgaard and Taksar (2001 [83]) in the
case of unbounded dividend rate. Here the corresponding HJB equation is
2 2
a + x2 I2 00 0 0
max max V (x) + (a + xrI )V (x) V (x) , 1 V (x) = 0
a[0,1] 2
The explicit form of the solution of (4.2.21) is rather complicated and usually
can be expressed through the special functions. For instance, in the case of
risk-free investment (i.e. I = 0) it can be shown (e.g. see Polyanin and
Zaitzev (2002 [119])) that the solution of (4.2.21) can be expressed in terms
of hypergeometric functions.
It was shown in Paulsen and Gjessing (1997 [117]) that in case of I > 0
the solution of (4.2.21) can be expressed as a linear combination of two
different special functions.
It can be shown (e.g. see Hjgaard and Taksar (2001 [83])) that if the
discount rate is less than rI , then the optimal value function V is infinite.
If = rI , then the value function V (x) is finite for all x < , however no
optimal policy exists. If > rI , then there is a finite level x1 > 0, such
that the optimal dividend policy is to distribute all surplus exceeding x1
as dividends. At the same time there exists x0 < x1 , such that the optimal
insurers retention limit of the quota share proportional reinsurance increases
monotonically on (0, x0 ) from 0 to 1.
Chapter 4 151
In this particular case of stochastic control the dynamics (4.2.6) of the com-
panys surplus can be rewritten in the following way
dR0 = x. (4.2.24)
This problem was investigated in the papers by Taksar and Zhou (1998 [142])
and Choulli et al. (2003 [28]) in the case of an unbounded dividend rate.
Here the corresponding HJB equation is
2 2
a 00 0 0
max max V (x) + (a )V (x) V (x) , 1 V (x) = 0
a[0,1] 2
The solution to this HJB equation varies in the following three cases.
1) If then V (x) = x, L (t) = x for all t 0 and the ruin time
is zero under the optimal control policy. In other words, if the expected
per unit time premium income is less than the rate of liabilities, then the
optimal policy is to declare bankruptcy immediately and distribute all the
companys surplus among shareholders as dividends.
2) If < 2 then
(
C e+ ()x e ()x , x x1 ;
V (x) =
C e+ ()x1 e ()x1 + x x1 , x > x1 ;
()
ln ()
where x1 = 2 + ()
+
()
and the constant C can be defined from the
smoothness conditions. The optimal dividend policy is the barrier policy
with barrier x1 , and the optimal risk control policy is not to buy reinsurance.
3) If 2 < then the optimal value function has the following form
x
R 1
G (y) dy, x < x0
0 C1 ,C 2
V (x) = C3
e + ()(xx0 ) C4
() e ()(xx0 )
, x 0 x < x1 ;
+ ()
x x1 + C 5 , x x1 ;
Chapter 4 152
2
1 2 2 ln y
where GC1 ,C2 (y) = C1 y + C2 2
, and Ci , i = 1, ..., 5, x0 and x1
+
2 2
can be found using smooth-pasting conditions.
The optimal dividend policy is the barrier policy with barrier x1 . The
optimal insurers retention level of the quota share proportional reinsurance
is
1 0
1
a (x) = GC ,C (x) GC ,C G C ,C (x) ,
2 1 2 1 2 1 2
In the previous subsection for the different diffusion models of the companys
surplus we found the optimal dividend and risk control policy under which
the expected present value of future dividends (value function) attains its
maximum, and this optimal dividend policy is a barrier policy with barrier
x1 . Whenever the companys surplus exceeds this reflecting point x1 the
excess is paid out to shareholders as dividends. It means that under the
optimal dividend policy the value of the surplus net of dividends (state
variable x) is always between 0 and x1 . As can be seen from the derivation
of the optimal dividend policy the barrier x1 has the following property: it
is a switching point at which V 0 (x1 ) = 1 and due to the concavity of the
value function V (see Proposition 4.2.1) V 0 (x) > 1 for all x (0, x1 ).
For x (0, x1 ), let us consider the following ratio
V (x)
R(x) = ,
x
which represents the expected present value of future dividends per unit
V (x)
of initial surplus (the risk capital supplied by shareholders). Since x
=
Chapter 4 153
V (x)V (0)
x0
, the ratio R is the slope of a secant. Taking into account the
concavity of V we conclude that R(x) is a decreasing function and for all
x (0, x1 )
This means that V (x) > x, that is for all possible risk-free discount rates
and for all possible initial value of surplus x net of dividend payments
the optimal value function (expected present value of future dividends) is
always greater than the initial surplus x. This is a somewhat strange result
since there is no such risk-free discount rate under which V (x) = x. This
can be explained in the following way. The expected present value of future
dividends is calculated under the physical (historical) probability measure
which does not capture all risk factors. That is why the rate of return
on equity is not adjusted for the risk, i.e. it is lower and thus V (x) >
x. To find the fair risk-adjusted rate of return one has to find such risk-
adjusted probability measure under which the expected present value of
future dividends is equal to the initial surplus.
It is worth noticing that the fair risk-adjusted rate of return is a very
important measure of investment performance for the shareholders. Indeed,
if the shareholders, as investors in the capital market, would invest all of their
wealth in one particular company, then the maximization of the expected
value of future discounted dividend cash flow is an appropriate objective for
the shareholders. However, it is known from CAPM theory that investors
behave rationally and do not have a large part of their wealth tied up in
one particular security, i.e. they diversify their investment risk through
purchasing optimal amounts of securities that provide fairly high return.
That is why we consider shareholders as companys investors that make
their investment decisions based on a NPV-zero criterion, meaning that the
market value of the shareholders equity claim (expected present value of
Chapter 4 154
Z Z
1 1
x = V (x) = EP e t l (t) dt = EP e[+(x,t)] t l (t) dt
R(x) R(x)
0 0
Z
= EQ e t l (t) dt,
0
and Q is the risk-adjusted probability measure. The term (x, t) plays a spe-
cial role of risk loading on the risk-free discount rate . It has the following
properties:
barrier. The retention level of the companys underwriting risk is also given.
The value function is maximized by these two uncontrolled variables: the
insurers retention level of quota share proportional reinsurance and the
dividend barrier. Modelling of the companys surplus by general Ito diffusion
covers two major types of diffusion approximation used in modern research
literature:
1) approximation by arithmetic Brownian motion;
2) approximation by geometric Brownian motion.
One obtains the approximation by an arithmetic Brownian motion when
the classical risk model of the insurers surplus (Cramer-Lundberg model) is
approximated by a diffusion process. One can use an insurance economics
model, in which the companys surplus is decomposed into subordinate pro-
cesses for incurred losses, earned premiums. Approximating each of these
subordinate processes by a diffusion gives us an analytical form of the sur-
plus which is governed by a geometric Brownian motion (eg see Powers (1995
[121])).
R0 = x,
Consider a dividend policy with barrier b > 0, i.e. whenever the com-
panys surplus exceeds the level b the excess is paid to shareholders as div-
idends. The dividend barrier is usually treated as a reflection barrier. The
aggregate dividends paid until time t are
R0 = x
The value function is defined to be the present value of all dividends paid
until ruin time = inf{t : Rt = 0}:
R t
e dLt ; 0 x b;
V (x, b) = 0
x b + V (b, b), x > b
0 (x) , x 0.
g 00 (b ) = 0.
Examples
Rt = R0 + t + Wt Lt .
g(x) = e+ x e x ,
Chapter 4 160
2 +2 2
where = 2
. The value function is
g(x) e+ x e x
V (x, b) = = .
g 0 (b) + e+ b e b
ln | | ln +
b = 2 <
+
g(x) = C1 ( x + )1 + C2 ( x + )2 , x [0, b]
C2 = C1 1 2 ,
1
!
2 (2 1) 1 2
b = 1 ,
1 (1 1)
1 1
C1 = .
(2 1 )2 ( b + )2 1 1 2
1 2
+ I2 x2 V 00 (x, b) + ( + rI x) V 0 (x, b) V (x, b) = 0, 0 < x b.
2
It is easy to show that if 0 (x) = rI > (the drift condition is violated) then
the value function is infinite (eg. see Hjgaard and Taksar (2001 [83])). In
the case where rI < , Paulsen and Gjessing (1997 [117]) found a solution of
the latter ODE using the method of contour integration. The form of this
solution is quite complicated and is represented in terms of hypergeometric
functions.
The first obvious property of the value function is that V (x, b) is an in-
creasing function of the initial surplus x. This immediately follows from the
g(x)
analytical representation of V through the function g: V (x, b) = g 0 (b)
and
g 0 (x)
hence x
V (x, b) = g 0 (b)
> 0, since according to the part 1) of Proposition
4.3.1 g 0 (x) > 0 for all x.
Using part 2) of Proposition 4.3.1 we conclude that the value function
V (x, b) as a function of x is
concave on [0, b b ];
linear on [max{b, b }, ];
convex on [b , b] if b > b ;
From here we can see that the risk-adjusted discount factor is m(t) =
e(+ t )t < et . Now the time dependent risk loading ln R(x,b) on the
ln R(x,b)
t
Chapter 4 163
risk-free discount rate is higher when the initial surplus is lower (i.e. the
companys level of capitalization is lower), and vice versa. This is under-
standable, since whenever shareholders invest in an undercapitalized com-
pany they expect a higher rate of return to compensate the risk of the com-
panys insolvency with respect to which they are sensitive. Note that such
insolvency risk is reflected in the R or in the time dependent risk loading on
the risk-free discount rate .
If we consider the initial surplus to be x = b , then V 00 (b , b ) = 0,
V 0 (b , b ) = 1 and thus using ODE that the value function satisfies we obtain
(b )
V (b , b ) = .
(b )
One can notice that b < V (b , b ) =
. In the case where and are
constants V (b , b ) is independent of . This may happen only if the optimal
dividend barrier b is such function of that V (b (), b ()) =
. In this
particular case it is possible to show (see Gerber and Shiu (2004 [67])), using
parametrization of the analytical form of b (, , ), that b is an increasing
function of ; b
as and b 0 as 0.
So far we have analyzed the value function V (x, b) for the Ito diffusion
model of the companys surplus with initial surplus x and the dividend
barrier policy with barrier b. We know that for every x 0 there exists
b (finite or infinite) under which V takes its maximum value. Below we
investigate the function V when x = b (i.e. function v(b ) = V (b , b )) and
provide some properties.
(b)
v(b) () .
Chapter 4 164
Proof. Note that for all b > 0 the function v(b) satisfies the following equa-
tion
1 2
(b) v00 (b) + (b) v(b) = 0,
2
00 2
where v (b) = x2 V (x, b) and v0 (b) = x
V (x, b)x=b = 1 (boundary
x=b
condition that holds for any barrier b).
(b) 1
Hence v(b) =
+ 2
2 (b) v00 (b).
g 00 (b )
Since b is an optimal barrier then v00 (b ) = g 0 (b )
= 0.
Now using part 2) of Proposition 4.3.1 we conclude that for b > (<)b
we have v00 (b) ()0.
Remark. This result has the following implication in the case where
the drift and diffusion coefficients are constant. We consider the situation
when the company, which uses the dividend barrier policy with barrier b, is
just about to continue its business after paying dividends. In this situation
the expected present value of dividend payments until ruin time is v(b).
We know that in this particular case there exists a finite optimal barrier
b such that V (x, b ) is maximal for all x > 0 (b is independent of x).
Then according to the latter Proposition for any two barrier policies such
that b1 < b < b2 we have v(b1 ) <
< v(b2 ). This means that if the
insurer increases (decreases) the dividend barrier with respect to the level
b the expected present value of future dividends where the initial surplus
coincides with the dividend barrier (i.e. the function v(b)) will increase
(decrease). This is opposite to the intuitive expectations where one may
expect that lowering the dividend barrier will increase the size of dividend
payments. The result of the latter Proposition refutes this view and it is
due to the taking into consideration the effect on the ruin time of changing
the barrier.
Chapter 4 165
2
b (a) = a ln = a b .
+ +
The value function under the optimal dividend barrier, initial surplus x <
b (a) and quota share proportional reinsurance with retention level a (0, 1)
is equal to
+
x x
e a e a 1 g(kx) V (kx, b )
Va (x, b (a)) = + = =
+
e a
b (a)
e a
b (a) k g 0 (b ) k
a a
b
= x R(kx, b ), 0 x ,
k
where k = a1 . Note that the inequality x b (a) = bk for any fixed x and
b (x < b ) implies that k 1, bx . As was established in the preceding
subsection the ratio R(kx, b ) is a decreasing function of k on k 1, bx .
This means that the value function Va (x, b (a)) takes its maximum at k = 1
or equivalently at a = 1. So, this means that reinsurance is reluctant in this
case, i.e. it is optimal not to take reinsurance if one wants to maximize the
value function.
On the other hand if one is interested in the risk-adjusted discount rate
Chapter 4 166
then taking quota share proportional reinsurance will decrease the value
Va (x,b (a))
function Va (x, b (a)), or equivalently, will decrease the ratio x
=
ln R(kx,b )
R(kx, b ) or the same as the risk loading t
on the risk-free discount
factor . This is understandable, since taking reinsurance will increase the
companys solvency strength which will cause the decrease in risk loading
on the risk-free discount factor.
say b, under which the companys ruin probability over the finite period of
time [0, T ] does not exceed a certain level > 0. It is clear that under
such solvency restrictions no dividends are allowed to be paid out unless
the surplus is greater than b. The main question that can be raised here
is regarding the optimal dividend barrier under the presence of solvency
restrictions if b > b . In subsection 4.4.1 we attempt to answer this question
and present one of the standard methods of calculation of the minimal level
b of risk capital.
In subsection 4.4.2 we show that the maximization of expected present
value of future dividends under solvency restrictions defined by ruin proba-
bility is approximately equivalent to the maximization of expected present
value of certain type of isoelastic utility of dividend payments. Using util-
ity maximization approach we consider some special type of uncontrolled
models of the surplus which are governed by geometric Brownian motions,
and for which the shareholders value is optimized by uncontrolled variables:
leverage ratio (insurance supply index), retention level of quota share pro-
portional reinsurance and dividend payment rate (rate of proportionality to
the surplus). In addition to this the insolvency is defined by a regulators
boundary for which the company becomes insolvent whenever the surplus
falls below this boundary, and when the company is insolvent it experiences
the insolvency costs.
Chapter 4 168
R0 = x
In the preceding section it was shown that the value function which is defined
to be the present value of all dividends paid until ruin time = (x, b) =
inf{t : Rt = 0 | R0 = x}:
R
et dLt ; 0 x b;
V (x, b) = 0
x b + V (b, b), x > b
g(x)
V (x, b) = ,
g 0 (b)
1 2
(x) g 00 (x) + (x)g 0 (x) g(x) = 0, x 0.
2
The optimal dividend barrier b under which the value function V attains
its maximum is defined by the equation g 00 (b ) = 0.
Now consider the regulators boundary b which represents a minimal level
of risk capital (initial surplus) with which the companys ruin probability
on finite period of time [0, T ] does not exceed a pre-specified level > 0.
Chapter 4 169
In other words the regulators boundary can be found from the equation of
ruin probability
= P (x, b) < T = x, T ; b .
It is clear that the company is not allowed to pay out dividends unless
the surplus exceeds the boundary b. Therefore, the regulators boundary b
can be treated as another dividend barrier for which we get the following
restriction on the dividend payments
Z
1{Rs <b} dLt = 0.
0
Now all dividend-barrier policies with the barrier b are defined to be admis-
sible under the solvency constraints if b b. In the case where the optimal
dividend barrier b , obtained in the previous section, is greater than the
regulatorys boundary b, the dividend policy with the barrier b is admissi-
ble and thus b remains the optimal dividend barrier. However, in the case
where b < b the question as what is then the optimal dividend barrier is
not trivial. The following proposition gives us an answer to this question.
Proof. We apply the Itos formula for the jump processes (eg. see Schonbucher
(2003 [130]) and Jacod and Shiryaev (1988 [87])) to the function et V (Rt , b):
for any admissible dividend-barrier policy with the barrier b > b and t 0
Chapter 4 170
we have
e (t (x,b)) V (Rt (x,b) , b) = V x, b
Z(x,b)
t
s 1 2 00
0
+ e (Rs ) V Rs , b + (Rs ) V Rs , b V Rs , b ds
2
0
Z(x,b)
t Z(x,b)
t
s 0
+ e (Rs ) V Rs , b dWs e s V 0 Rs , b dLcs
0 0
X
+ V Rs , b V Rs , b V 0 Rs , b Rs
st (x,b)
Z(x,b)
t
= V (x, b) + e s (Rs ) Rs b + V (b, b) 1{Rs >b} ds
0
Z(x,b)
t Z(x,b)
t
+ e s (Rs ) V 0
Rs , b dWs e s V 0 Rs , b dLs
0 0
X
s
+ e V Rs , b V Rs , b ,
st (x,b)
e (t (x,b)) V (Rt , b) = V x, b
Z(x,b)
t
+ e s (Rs ) Rs b + V (b, b) 1{Rs >b} ds
0
Z(x,b)
t Z(x,b)
t
+ e s (Rs ) V 0
Rs , b dWs e s dLs .
0 0
Using the assumption 0 (x) for which there exists a twice continuously
differentiable function V , we obtain for all x b
which is equivalent to
According to Shreve et al (1984 [137]) V 0 (x, b) is bounded on [0, b] for (x) >
0. It follows from this that
t (x,b)
Z
E e s (Rs ) V 0 Rs , b dWs = 0.
0
Hence,
t (x,b)
Z
0 E e (t (x,b)) V (Rt , b) V x, b E e s dLs ,
0
or equivalently when t
(x,b)
Z
V x, b E e s dLs .
0
From this proposition we see that if the truncated (by the boundary b)
set of all admissible dividend-barrier policies such that their barrier b b
does not contain b = b , then the optimal dividend barrier is b = b.
Finally the regulators boundary b can be found from the initial solvency
restriction imposed by a regulatory:
the survival probability of the insurer over the finite period [0, T ] with the
initial minimal level of capitalization b is greater or equal to 1 with
0 < 1, i.e.
(x, T ; b)|x=b = 1 (x, T ; b)|x=b = 1 .
differentiable w.r.t t satisfies the following second order PDE for all x [0, b]
and t > 0
1 2
(x, t; b) = 2 (x) 2 (x, t; b) + (x) (x, t; b) (4.4.1)
t 2 x x
(x, 0; b) = 1, 0 < x b;
(0, t; b) = 0, (x, t; b) = 0, t > 0.
x
Indeed, one can apply Ito formula for the jump processes to the function
Rt (x,b) , T (t (x, b)) ; b , 0 < x b and obtain
Rt (x,b) , T (t (x, b)) ; b = (x, T ; b)
Z(x,b)
t
1 2 2
+ (x) 2 (Rs , T s; b) + (x) (Rs , T s; b)
2 x x
0
Z(x,b)
t
(Rs , T s; b) ds + (Rs ) (Rs , T s; b) dWs
t x
0
Z(x,b)
t
(Rs , T s; b) dLs = (x, T ; b).
x
0
All integrals in the latter equality are equal to zero due to the fact that the
function satisfies PDE (4.4.1) with its boundary conditions. Now, taking
expectation from both sides of the latter equality under t = T we get
(x, T ; b) = E Rt (x,b) , T (t (x, b)) ; b
= E (RT , 0; b) 1{T < (x,b)} + E (0, T (x, b); b) 1{T (x,b)}
= E 1{T < (x,b)} = P [T < (x, b)] ,
and we can see that the function that satisfies PDE (4.4.1) is indeed a
survival probability function of the insurer. To solve PDE (4.4.1) one may
Chapter 4 173
intention to maximize its shareholders value usually shifts the insurer to the
state with a higher leverage level, which is achievable by issuing more in-
surance debt (or by assuming more underwriting risk). But at the same
time assuming more underwriting risk will require the insurer to be more
financially reliable. And this financial reliability (strength) is controlled by
a regulator via imposing some specific constraints on underwriting risk, such
as for example maximally acceptable levels of ruin (insolvency) probability
or volatility. Regulators impose solvency requirements in order to protect
insureds who accumulate their insurance credit exposure with one or few
insurers and thus are sensitive to the insolvency risk.
Continuing this logic further we argue that the insurer should consider
some preference value as a function of its wealth, which is associated with
the fixed level of insolvency risk. This preference value should be higher for
a larger monetary amount of wealth, and that the gain of the preference
value resulting from the same monetary gain is a decreasing function of ini-
tial wealth. In other words the preference value is an increasing function of
wealth and its marginal function is a decreasing function of wealth, i.e. the
preference value represents a concave utility function of the insurer (i.e. the
insurer is risk-averse). It is worth noticing that this analysis is consistent
with the fact that the mean-variance criterion can be reconciled with the ex-
pected utility approach using a quadratic utility function (see eg Markowitz
(1959 [100])).
In this subsection we will show that the maximization of shareholders
value under solvency restrictions is approximately equivalent to the maxi-
mization of shareholders value using utility approach with a specific isoelastic
utility function.
Using this isoelastic utility function we maximize the expected present
value of utility of future dividends by three uncontrolled variables: dividend
Chapter 4 175
where the annual total claims St , t = 1, 2, ... are independent and iden-
tically compound Poisson distributed, say St S; the initial surplus R0
is equal x and c is the annual premium. There exists a minimum annual
premium c = P such that for the fixed initial surplus x the insurers ruin
probability attains its maximally acceptable level defined by a regulator.
We may assume that this level can be approximated by Lunberg upper bound
ex , where denotes the adjustment coefficient and is thus the root of the
equation ec = E eS . Hence,
1 S
P = ln E e ,
|ln |
where = x
. From practice we now that 0 < 1 and therefore using
Taylors expansion we get
S 1 2 2 2
ln E e = ln 1 + E [S] + E S + o( )
2
2
1 2 2 2 1 1 2 2
= E [S] + E S + o( ) E [S] + E S + o( ) + o(2 )
2
2 2 2
1
1 2
= E [S] + E S 2 2 + o(2 ) E [S] 2 + o(2 ) + o(2 )
2 2
1 2 2
= E [S] + Var [S] + o( ).
2
Chapter 4 176
From here we conclude that the minimal insurers annual premium rate is
1
P = E [S] + Var [S] + o()
2
and hence it can be approximated by
|ln |
P E [S] + Var [S] .
2x
On the other hand the same minimal insurers annual rate P can be cal-
culated using expected utility approach for the period from 0 to 1. Let the
unknown insurers utility function is U , then P must satisfy the following
equation
E U x + P S = U (x). (4.4.2)
The fact that the insurer is risk-averse implies that the marginal utility U 0
is positive decreasing function. Therefore, the latter equation determines
P uniquely, but has no explicit solution in general. However, using the
parametrization2 of S : S(t) = E[S] + tY , with E[Y ] = 0, and expansion of
P
P in powers of t, i.e. P (t) = pk tk , we can approximate P by the
k=0
mean of S plus variance of S with coefficient of proportionality determined
00
by Arrow-Pratt absolute risk aversion coefficient r(x) = UU 0 (x)
(x)
.
In order to find the first three terms in series of P (t) we first set t = 0
in (4.4.2) and obtain
U (x) = U x + P (0) S(0) = U (x + p0 E[S]) ,
E [(p1 Y ) U 0 (x)] = 0,
2
This idea is borrowed from Gerber and Pafumi (1998 [66]) who approximated the
certainty equivalent gain of the decision-maker.
Chapter 4 177
0 = E [2p2 U 0 (x)] + E (p1 Y )2 U 00 (x)
1
P (t) = E[S(t)] + r(x) Var [Y ] t2 + o(t2 )
2
1
= E[S(t)] + r(x) Var [S(t)] + o(t2 )
2
1
P E[S] + r(x) Var [S] .
2
|ln |
r(x) = ,
x
which means that the insurers utility function must satisfy the following
ODE
|ln | 0
U 00 (x) + U (x) = 0.
x
x1m
This ODE gives us the solution U (x) = 1m
with m = |ln | > 0. Note that
m 6= 1 for all reasonable values of < 10%.
So, we conclude that the corresponding insurers utility function has the
constant relative risk aversion (CRRA) coefficient 0 < m 6= 1 and thus this
utility function is isoelastic.
Chapter 4 178
We consider the diffusion model of the surplus from the RBC solvency model
of the insurer introduced in Chapter 3. That is, the surplus is described by
a geometric Brownian motion
- leverage ratio;
be declared insolvent if the surplus falls below this boundary. Let there also
be an insolvency cost K. The insolvency cost may include payments made
to third parties other than policyholders and shareholders, as well as legal
fees and other costs (see Rajani (2002 [122])). From the regulators point
of view, the insolvency cost may be defined to include both policyholders
and shareholders losses associated with the insolvency event. At the time of
insolvency = inf {t : Rt = x } the residual surplus net of insolvency cost
is distributed among shareholders as terminal dividends.
Here we define the value function (shareholders value) by
Z
V (x) = V (x; a, , ) = E es U (ls ) ds + e U (K)
0
x1m
with the boundary condition V (x ) = U (K) for U (x) = 1m
with m =
|ln | > 0, and where the dynamics of the surplus are defined by the diffusion
model (4.4.3). The important property of V that determines the second
boundary condition is that V is bounded.
The corresponding HJB equation which the value function must satisfy
is
1 2 2 00
x V (x) + ( ) x V 0 (x) V (x) + U (x) = 0, x x
2
subject to the boundary condition V (x ) = U (K). One can note that the
latter HJB equation is a Cauchy-Euler second order ODE, which can be
solved using the transformation of the variable x = ey . Doing so, we get
1 2 00 1 2 0 (ey )1m
Vyy + Vy V = .
2 2 1m
The general solution to the latter ODE is
V (y; a, , ) = C e y + C+ e+ y + Vb (y; a, , ),
q
2
( 12 2 ) ( 21 2 ) +2 2
where = 2
, and Vb (y; a, , ) is the partial
solution. By introducing the partial solution in the form Vb (y; a, , ) =
Chapter 4 180
or equivalently
1
1m 1 2 2 1 2
g(a, , ) = (1 m) + (1 m) .
1m 2 2
It should be noticed that < 0, + > 0 and 1 m < 0 for < 10%. There-
fore, when x and thus when y the term C e y is unbounded,
implying that C = 0.
Hence, putting = + the solution to the HJB equation becomes
K 1m
Finally, imposing the first boundary condition, V (x ) = 1m
, gives us
K 1m
C+ = (x ) g(a, , ) (x )+1m ,
1m
and thus
K 1m 1m x
V (x; a, , ) = g(a, , ) (x )
+ g(a, , ) x1m .
1m x
One can notice that the optimal leverage, reinsurance and dividend rate
are, in particular, dependent of the insolvency cost. It is well-known that
in corporate financial theory, one possible reason for the existence of an
optimal leverage is the bankruptcy cost. Likewise, in the insurance model
Chapter 4 181
considered above, the insolvency cost can be a factor that leads to an optimal
leverage and especially to an optimal reinsurance. As to optimal reinsurance
strategy, it has been shown in the previous subsection that without solvency
cost, which may be treated as frictional costs, it is optimal not to buy
reinsurance. Here the presence of insolvency cost may induce the demand
for reinsurance in maximization of shareholders value.
4.5 Conclusion
This chapter considered the problems of reinsurance and dividend optimiza-
tion in a dynamic setting. It started from an analysis of different stochastic
control models of dynamic insurance risk management, namely models in
which the expected value of future, until ruin time, discounted dividend
payments are controlled via two control variables: dividend and reinsurance
policies. We showed that these models generate a demand for reinsurance
and that the optimal value function, defined as the expected present value of
future dividends, per unit of initial surplus is always greater than one under
a physical probability measure and a constant risk-free discount rate. We
found a risk-adjusted probability measure under which the value function is
equal to the initial surplus and showed that the corresponding risk loading
on the risk-free rate of return is a decreasing function of the insurers level
of capitalization (initial surplus).
We have also investigated the expected present value of future dividends
(shareholders value) in general Ito diffusion models of a companys surplus
with dividend barrier strategies and analyzed the demand for reinsurance in
these uncontrolled models. We also investigated the properties of optimal
dividend barrier and again showed that the risk-adjusted discount rate is
a decreasing function of the insurers surplus. We reconsidered the same
Chapter 4 182
models under the solvency restrictions and showed that there is a trade-off
between the reinsurance and optimal dividend barrier. Finally, we rein-
vestigated these problems under the presence of insolvency costs using an
expected utility approach and derived the optimal value function.
Chapter 5
Competitive equilibrium
pricing in an integrated
insurance-reinsurance market
183
Chapter 5 184
5.1 Introduction
The equilibrium studies in insurance and in economics in general were orig-
inated by Arrow (1953 [4], 1954 [6]) and Debreu (1953 [42]), who used con-
tingent space to study economic equilibrium in a simple risk exchange model
consisting of two risk averse agents. They showed that competitive equilib-
rium is Pareto optimal, and every Pareto optimal strategy can be supported
by a competitive equilibrium through the redistribution of endowments, in
other words they proved that the first and the second social welfare theorems
hold in an economy with uncertainty.
Later Borch, inspired by this result of Arrow and Debreu, used the foun-
dations of general equilibrium theory to tackle the problems of optimal risk
sharing in insurance. Borchs key results, published in the early 1960s (1962
[16]) and later improved by Du Mouchel (1968 [51]), provide with the con-
dition of Pareto optimal risk sharing between several insurance companies.
This result is very important from an insurance economics point of view
and establishes the notion of Pareto optimality in an insurance market, or
a special equilibrium under which it is impossible to make one agent better
off without making an other agent worse off.
Many studies of insurance market equilibrium consider a single period
setting and assume either that the price of insurance is given exogenously
or that the quantity of insurance purchased is fixed. On the demand side,
the papers by Gould (1969 [71]), Mossin (1968 [107]), Smith (1968 [138]),
Ehrlich and Becker (1972 [52]), Raviv (1979 [123]) and Moffet (1979 [105])
have assumed that the price of insurance is exogenous, implying the implicit
assumption that the demand curve has no effect on market price. On the
supply side, for instance, works of Biger and Kahane (1978 [12]) and Fairley
(1979 [55]) used CAPM to derive the price of insurance, and subsequently,
Chapter 5 185
Doherty and Garven (1986 [49]) and Cummins (1988 [35]) used option pric-
ing models to determine insurance price. In both approaches, it was assumed
that the shift in the supply curve has no impact on market quantity. The
results of these studies of insurance demand and supply may lead to unfair
insurance price for either insurance shareholders or policyholders. Such price
may not be an equilibrium market price at which the responses of insureds
and insurers are reciprocally optimal.
Cummins and Sommer (1996 [38]) and Venezian (1994 [146]) indicated
that the insurance market price can also be affected by the cost of possible
insolvency. A lower leverage or a higher capitalization level, in general, will
reduce the probability of ruin, leading to a higher insurance premium. In
other words the insurance market price is directly affected by the quality of
the insurance products, associated with insolvency risk. One can consider
the economic asset-liability approach of evaluation of (re)insurance liabili-
ties and define a (re)insurance premium as the risk-neutral present value of
liabilities minus the value of so called insolvency exchange put option that
reflects the risk-neutral present value of deficiency when the insurer is insol-
vent (see Sherris (2004 [134]) or Grundl and Schmeiser (2002 [75])). Here
we also observe the effect of insolvency on equilibrium (re)insurance price: a
lower leverage or a higher capitalization will reduce the value of insolvency
exchange put option, which leads to an increase in the (re)insurance price.
Many traditional insurance or financial models are static and involve only
one-period objective function optimization. However, many decision-making
processes of the insurance firm are more appropriate for multi-period rather
than single-period models because most firms have multi-period planning
horizons. For example, the optimal multi-period dividend payout policy of
an insurance company to its shareholders may not be the same as that of
a single-period policy. Moreover, most insurers investments are long-term,
Chapter 5 186
(X) = E[X], X L2 .
The random variable is called the state price deflator. The problem each
agent intends to solve is the following
The proof of this theorem can be found in Aase (2002 [1]) and is built
on the Kuhn-Tucker theorem for functionals.
Pareto optimality
where there exists at least one i for which the inequality is strict.
It can be proved, using the standard separating hyperplane theorem, that
the risk sharing Y = (Y1 , Y2 , ..., YM ) is Pareto optimal if and only if there
exist positive weights 1 > 0, 2 > 0, ..., M > 0 such that Y solves the
following optimization problem
M
X
sup i E [ui (Zi )] ,
Zi L2
i=1
M
X M
X
subject to Zi Xi = W.
i=1 i=1
Chapter 5 189
P
M
where u is a market utility function defined on aggregate wealth W = Xi .
i=1
We can represent the state price deflator using the market utility function
in the following way
P
M
i u0i (Yi (W ))
u0 (W ) i=1
=
E [u0 (W )] P
M
i E [u0i (Yi (W ))]
i=1
PM
i (W ) E [u0i (Yi (W ))]
a.s. i=1
= = (W )
P
M
i E [u0i (Yi (W ))]
i=1
1
It follows from the fact that the price of payoff represented by a certain deterministic
constant is identical to this constant.
Chapter 5 190
agents of the market with Pareto risk sharing Y (W ) = (Y1 (W ), ..., YM (W )).
r (w)
Yi0 (w) =
ri (Yi (w))
2) the risk tolerance of the market is the sum of the risk tolerances of all
agents under Pareto optimal risk sharing, i.e.
XM
1 1
=
r (W ) r (Y (W ))
i=1 i i
The proof of the first part can be found in Buhlmann (1980 [26]) and the
proof of the second part can be found in Borch (1985 [18]).
The latter theorem is very important in a theory of risk sharing in an in-
tegrated insurance-reinsurance market. For instance we can apply the result
of the first part of this theorem to the problem of risk sharing in a simple
duopoly insured-insurer or insurer-reinsurer to get the Pareto optimal risk
sharing: insurance or reinsurance. Let us consider the first agent (an insured
or an insurer) with strictly concave utility function u1 (it is risk averse) and
initial capital w1 , and the second agent (an insurer or a reinsurer) with utility
function u2 and initial surplus w2 . These two agents can negotiate an insur-
ance contract, according to which the second agent pays the indemnity I(X)
to the first agent if the last one experiences losses X. Here the indemnity I
is treated as a transformation of X for which 0 I(X) X, I(0) = 0. Let
P denotes the premium for this contract. Then according to the first part of
the latter theorem the Pareto optimal risk transfer I satisfies the following
Chapter 5 191
and at the same time the smallest premium P2 for which the second agent
is willing to assume the risk IP2 is given by
From the point of view of financial theory the insurance liability can be
treated as the corporate debt that an insurer raises from its insureds. The
insurance is risky, since unlike other financial debts, for which the claim
amount and payback date are deterministic, insurance debt is random in
severity, frequency, and the time of loss settlement. From the risk perspec-
tive, in addition to the cost of expected losses, an insurer charges a risk
premium for its unavoidable, undiversifiable risk of losses. In addition, an
insurer should pay the cost of exploiting the insurance fund, generated by
paid insurance premiums, through investment in the security market. Hence,
the actual insurance premium is the cost of expected insurance losses plus
risk premiums minus cost of using the insurance fund, and in general is
greater than the cost of expected insurance losses.
On the other hand from the point of view of financial theory the insurance
contract designed for a single period is a special type of financial security
with payoff X and the end of the period. According to the CAPM theory
of asset pricing the price of a financial security with payoff satisfies the
following fundamental value equation (see Cochrane (2001 [29]))
= E[X],
where is the stochastic discount factor. The stochastic discount factor can
Chapter 5 193
u0 (c1 )
= 0 ,
u (c0 )
1
= E[] E[X] + Cov[, X] = E[X] + Cov[, X],
1+r
1
where we have used E[] = 1+r
, and where r is a risk-free interest rate.
Most classical financial securities pay off well in good times. Thus payoffs
co-vary negatively with the discount factor . In contrast traditional insur-
ance contracts pay off well in bad times (insurance events associated with
insurance loss). Hence, payoffs co-vary positively with the discount factor
, which explains in an alternative way why the insurance price is higher
than the expected present value of insurance losses.
Insurance CAPM
insurance debt that are modelled with diffusion processes. The insurance
CAPM model considers three classes of claimants: shareholders, financial
debt-holders, and policyholders, each with contingent claim on the insur-
ance company. The CAPM formula of risk-adjusted premium is calculated
for a general contingent claim as a function of assets and liabilities, and
the corresponding beta is a linear combination of betas associated with the
CAPM price of assets, financial debt and insurance debt. This is a funda-
mental formula of insurance CAPM. The insurance CAPM presumes that
an insurance market is a part of a financial capital market, i.e. an insurance
portfolio is market-efficient and the price is set to achieve the market equi-
librium. However, one may argue that in the insurance market, the portfolio
price is not only subject to the equilibrium of the capital asset market, but
also is subject to the equilibrium of supply and demand between the insurer
and insured. Suppose there were no information or transaction costs for an
insured to access the capital market in hedging its risk. In this case it would
be more reasonable to assume that the insurance price is determined solely
by the equilibrium of the financial capital market. However it is costly for an
insured to hedge, for instance, the risk of damaging its property through the
capital market in practice. Thus insurance firms still provide a more con-
venient way for individuals to hedge their risks, which cannot be diversified
in the capital market. This explains that the equilibrium insurance price is
defined from the prospective of closed-system insured-insurer, and is subject
to the equilibrium of the supply and demand for insurance. However, it is
worth noticing that the equilibrium of a capital asset market may have a
particular impact on the insurers supply curve and possibly on the insureds
demand curve, and will thus affect the equilibrium price of insurance.
Chapter 5 195
The main difference between classical actuarial models and solvency mod-
els of insurance pricing is that actuarial models do not reflect the effect of
insolvency on policy payoff, while in solvency models the evaluation of insur-
ance liabilities, viewed as corporate debt, is economic. It is worth noticing
that insolvency risk is incorporated in both models. In actuarial models the
insurance premiums are calculated from insurers prospective (supply side)
so that the risk loading is increasing in insolvency risk (ruin probability),
and are not affected by the expected cost of possible insolvency (expected
deficiency caused by insurers insolvency). On the other hand in solvency
models, the insurance price is the equilibrium price, which is defined by de-
mand for insurance from insureds side and supply function of insurance from
shareholders side. Here the demand function for insurance is decreasing in
insolvency risk.
As in Cummins and Sommer (1996 [38]) and Sherris (2004 [134]) we
use a single-period approach to illustrate the equilibrium pricing in solvency
models. We assume that all premiums are collected at the beginning of the
period and all insurance claims are paid at the end of the period. At the
beginning of the period the insurers assets A0 consist of premiums P0 and
risk capital (equity) E0 supplied by shareholders. All assets at time 0 are
invested in financial instruments with time-1-payoff A1 = (1 + rA )A0 , where
rA is the rate of investment return. The terminal value of insurance claims
(losses) is a random variable L1 .
The main economic (natural) assumption is to assume that an insurer
cannot pay insurance indemnities to its policyholders at the end of the pe-
riod at the level higher than the terminal value of its assets. This is the
assumption of the limited liability of the insurer against its policyholders.
At the end of the period the shareholders value (terminal equity value)
Chapter 5 196
is
(
A 1 L 1 , A1 > L 1
E1 =
0, A1 L1 ,
incorporate an investment risk and model it along with the insureds wealth
with geometric Brownian motions. Following Lin and Powers we write the
diffusion process of the insureds wealth as
where
BtY and BtW denote the standard Brownian motions associated with
the loss and investment return processes, respectively.
Now, let T denotes the length of the insureds finite planning horizon, and
let B(T, WT ) the insureds bequest amount at time T . Assuming that the
insured is the maximizer of its discounted lifetime utility of consumption,
we define the insureds value function as
T
Z
V (t, Wt ) = sup E ers u(Cs ) ds + erT u(B(T, WT ))
Ct 0,t [0,1]
t
Chapter 5 199
(pt 1)Wt
t = ,
(1 ) Y2 Yt
which is
These properties of optimal retention are consistent with those found for a
single-period model. However, there is no guarantee that, under other forms
of utility function, there will be a similarity of results for optimal retention
level in dynamic and static settings.
It is worth noticing that the insureds optimal price is greater than 1
since the insurance retention is positive. This is consistent with the result
of analysis of insurance price from the CAPM perspective, where due to the
Chapter 5 200
positiveness of the covariance term the price is higher than the expected
value of insurance payoff.
S0 = x
- leverage ratio;
1 2 2 00
x V (x) + ( x l) V 0 (x) r V (x) + U (l) = 0, x x
2
NP (1 t (p )) Yt = NI t (p , pre ) St ,
(5.3.2)
NI (1 a (p , pre )) t (p , pre ) St = NR re
t (pre ) Stre
Numerical example
5.4 Conclusion
In this chapter we considered and analyzed different models of insurance
pricing. In Section 5.2 we analyzed the classical theory of risk sharing in an
insurance-reinsurance market from the point of view of financial economics
theory and explained the relationship between financial (competitive) equi-
librium pricing and Pareto optimality of risk sharing. We also analyzed the
insurance pricing from CAPM perspective and explained the difference be-
tween CAPM price of insurance as a special type of financial security and
CAPM price of traditional financial instruments. As one of the most impor-
tant approaches of economic evaluation of insurance liabilities we considered
a solvency model of insurance pricing, explained how the equilibrium price
of insurance is affected by the value of expected insolvency cost (the value of
insolvency exchange put option) and compared it with the classical actuarial
models of insurance pricing.
In Section 5.3 we considered a dynamic model of risk exchange between
insureds, insurers and reinsurers. In this model we assume that the mar-
ket consists of homogeneous insureds, insurers and reinsures. We model an
insureds wealth and surpluses of an insurer and a reinsurer with geomet-
ric Brownian motions. Assuming that all agents are utility maximizers we
3
Clearly insureds are willing to buy less insurance from insurers with low financial
strength. Also they are willing to pay less for a low quality insurance. Therefore insurance
can be treated as a Giffen product for which a decrease in the price brings about a
decreased demand.
Chapter 5 206
define the corresponding objective functions and derive the HJB equations,
which these value functions must satisfy. We consider a numerical example
and compute demand-supply curves of insurance and reinsurance. In this
numerical example we found equilibrium prices of insurance and reinsurance
and showed the presence of reinsurance market increases the equilibrium in-
surance price. We explain this effect by increase in the financial strength of
insurer after introducing the reinsurance market.
Chapter 6
Conclusion
This dissertation covered four topics of insurer risk management. The first
part concerned the problems of insurers optimal risk sharing between an
insurer and a reinsurer under mean-variance criterion. It aimed to general-
ize some problems of reinsurance optimization from the classical actuarial
literature and consider the problems of finding optimal reinsurance endoge-
nously using methods of convex programming. By considering a reinsurance
contract as a transformation of the underlying loss, and recognizing ex-
pected profit net of reinsurance and the variance of retained underwriting
risk as convex functionals of the reinsurance transformation, we provided
some new theoretical results regarding endogenous reinsurance optimization
for different convex premium principles. We showed that the form of op-
timal reinsurance essentially depends on the analytical form of reinsurance
premium principle. In particular, we showed that a stop-loss reinsurance is
optimal only for the mean value reinsurance premium principle, and that
for other forms of reinsurance premium principle an optimal reinsurance
belongs to the class of change-loss reinsurance (mixture of quota share pro-
portional and stop-loss reinsurance). Using these results we constructed the
conditions of existence for a reciprocally optimal reinsurance contract. We
207
Chapter 6 208
barrier strategies and analyzed the demand for reinsurance in these uncon-
trolled models. In these uncontrolled models we investigated the properties
of optimal dividend barrier and again showed that the risk-adjusted dis-
count rate is a decreasing function of the insurers surplus. We reconsidered
the same models under the solvency restrictions and showed that there is
a trade-off between the reinsurance and optimal dividend barrier. Finally,
we reinvestigated these problems in the presence of insolvency costs using
utility approach and derived the optimal value function.
The results of the second and third parts are very important and show
that the decision to reinsure can be treated as both a risk-management and
a capital-structure tool in shareholders value creation.
In the fourth part we investigated models of equilibrium pricing in an
integrated insurance-reinsurance market consisting of insureds, insurers and
reinsurers. We started from an analysis of static equilibrium models of in-
surance equilibrium pricing and then considered a dynamic model of optimal
risk sharing. In the case of the static models we analyzed the classical theory
of risk sharing in insurance-reinsurance market from the point of view of the
financial economics theory and explained the relationship between financial
(competitive) equilibrium pricing and Pareto optimality of risk sharing. We
also analyzed the insurance pricing from CAPM perspective and explained
the difference between CAPM price of insurance as a special type of financial
security and CAPM price of traditional financial instruments. As one of the
most important approaches of economic evaluation of insurance liabilities we
considered a solvency model of insurance pricing, explained how the equilib-
rium price of insurance is affected by the value of expected insolvency cost
(the value of the insolvency exchange put option) and compared it with the
classical actuarial models of insurance pricing. In the case of the dynamic
model we considered the insurance and reinsurance prices to be Walrasian
Chapter 6 211
equilibrium prices, i.e. prices under which the aggregate demand for and
supply of (re)insurance are equal. We described methods of finding such op-
timal prices and using a numerical example, we illustrated that the presence
of reinsurance market increases the equilibrium insurance price. We explain
such effect of reinsurance by increase in the financial strength of insurer after
introducing the reinsurance market.
Bibliography
[5] Arrow, K.J. Uncertainty and the welfare economics of medical care.
American Economic Review, Vol. LIII:pp. 941973, December 1963.
212
Bibliography 213
[20] Bowers, N., Gerber, H., Hickman, J., Jones, D. and Nes-
bitt, C. Actuarial mathematics. Society of Actuaries, 2nd ed.
Schaumburg, Ill., 1997.
[33] Cramer, H. Collective risk theory, a survey of the theory from the
point of view of the theory of stochastic processes. Skand. Jubilee Vol-
ume, Stockholm, 1955.
[34] Culp, C. The ART of Risk Management. John Wiley and Sons, 2002.
[40] Davis, M.H.A. Markov models and optimization. Chapman & Hall,
London, 1993.
Bibliography 216
[43] de Finetti, B. Il problema dei pieni. Giorn. Ist. Ital. Attuari, Vol.
11:pp. 188, 1940.
[46] Diamond, D.W. Debt maturity structure and liquidity risk. Quaterly
Journal of Econimics, Vol. 106:pp. 709737, 1991.
[53] Emanuel, D.C., Harrison, J.M. and Taylor, A.J. Diffusion ap-
proximation for the ruin probability with compounding assets. Scan-
dinavian Actuarial Journal, Vol. 58:pp. 240247, 1975.
[71] Gould, J.P. The expected utility hypothesis and the selection of
optimal deductibles for a given insurance policy. Journal of Business,
Vol. 42:pp. 143151, 1969.
Bibliography 219
[80] Hipp, C. and Taksar, M. Stochastic control for optimal new busi-
ness. Insurance: Mathematics and Economics, Vol. 26:pp. 185192,
2000.
[105] Moffet, D. The risk sharing problem. Geneva Papers on Risk and
Insurance, Vol. 11:pp. 513, 1979.
[109] Myers, S. and Cohn, R. Insurance rate regulation and the capi-
tal asset pricing model. In D. Cummins and S. Harrington, editors,
Fair Rate of Return in Property-Liability Insurance. Kluwer Academic
Publishers, Norwell, MA, 1987.
[122] Rajani, S. Insolvency costs and fees. Tolley Publishing Co, 2002.
[131] Sethi, S.P. When does the share price equal the present value of
future dividends? A modified dividend approach. Econ.Theory, Vol.
8:pp. 307319, 1996.
[137] Shreve, S.E., Lehoczky, J.P. and Gaver, D.P. . Optimal con-
sumption for general diffusions with absorbing and reflecting barriers.
SIAM Journal of Control Optimization, Vol. 22:pp. 5575, 1984.
[142] Taksar, M. Optimal risk and divident control for a company with
a debt liability. Insurance: Mathematics and Economics, Vol. 22:pp.
105122, 1998.
[143] Taksar, M. Optimal risk and divident distribution control models for
an insurance company. Math.Meth.Oper.Res., Vol. 1:pp. 142, 2000.