THE UNIVERSITY OF NEW SOUTH WALES

FACULTY OF COMMERCE AND ECONOMICS

SCHOOL OF ACTUARIAL STUDIES

INSURER RISK MANAGEMENT AND

OPTIMAL REINSURANCE

Yuriy Krvavych

A THESIS SUBMITTED IN FULFILLMENT OF THE

REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

January 2005
To my beloved wife Olya
Abstract

In finance the existence of corporate risk management is due to imperfections in

financial markets. One of the main imperfections is associated with the cost of cor-
porate risk that firms assume. Costly corporate risk creates a set of frictional costs
and thereby decreases corporate value. Financial corporations manage their risk
to reduce the expected value of frictional costs and enhance shareholders value,
and do so using a wide variety of tools. This dissertation primarily considers an
insurance company as a special type of financial corporation leveraged by risky
debt, and investigates the existence of risk management incentives in insurance in
the presence of frictional costs such as financial distress costs and costs caused by
the convexity of the corporate tax rate. Here one of the main tool of risk hedging
is reinsurance, a classical tool for risk transfer in insurance, and this dissertation
investigates demand for reinsurance in insurer value creation. Insurer risk man-
agement problems are also investigated here in a dynamic setting, where the main
objective is to find optimal reinsurance and dividend payments under which the
expected present value of future dividends is maximised. This dissertation also
generalizes some classic actuarial results of reinsurance optimization under the
mean-variance criterion. In this work optimal reinsurance is found endogenously
for different reinsurance premium principle using standard methods of convex
analysis. Finally this work considers an integrated market consisting of insureds,
insurers and reinsurers, and studies the effect of the presence of reinsurance in
this market on insurance price.

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.

Sydney, Australia Yuriy Krvavych

January 10, 2005

1
The research was funded by the International Postgraduate Research Scholarship
(IPRS)

vi
Table of Contents

Abstract v

Acknowledgements vi

Table of Contents vii

List of Figures x

Notations and Abbreviations xii

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

2 Finding an optimal form of reinsurance contract under mean-

variance criterion 10
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Convex nonlinear programming methods in finding an opti-
mal reinsurance contract . . . . . . . . . . . . . . . . . . . . 16
2.2.1 Reinsurance optimization for convex premium principles 17
2.2.2 Analysis of reinsurance optimization under mean value,
generalized dispersion and exponential reinsurance pre-
mium principles . . . . . . . . . . . . . . . . . . . . . 28
2.3 Reinsurance optimization in the case of non-convex premium
principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4 Imperfectly competitive equilibrium in a duopoly insurer -
reinsurer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.4.1 Mean value premium principle . . . . . . . . . . . . . 44
2.4.2 Standard deviation premium principle . . . . . . . . 46

vii
 viii

2.4.3 Esscher premium principle . . . . . . . . . . . . . . . 47

2.5 Convex reinsurance optimization
sures of risk . . . . . . . . . . . .
2.6 Optimal reinsurance under utility
mium principle . . . . . . . . . .
2.7 Conclusion . . . . . . . . . . . . .
under more general mea-
. . . . . . . . . . . . . . . 53
approach and convex pre-
. . . . . . . . . . . . . . . 55
. . . . . . . . . . . . . . . 59

3 Enhancing insurer value via reinsurance optimization: in-

surance economics approach in a single period model 61
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.2 Demand for reinsurance in shareholders value creation: one-
period frictionless model . . . . . . . . . . . . . . . . . . . . 66
3.2.1 Maximizing return on risk capital by reinsurance and
risk capital . . . . . . . . . . . . . . . . . . . . . . . 69
3.2.2 Maximizing return on risk capital by reinsurance . . 77
3.3 Demand for reinsurance in shareholders value creation: one-
period model in the presence of corporate tax . . . . . . . . 87
3.4 Demand for reinsurance in shareholders value creation: one-
period model in the presence of costs of financial distress . . 100
3.4.1 Models setup and definitions: SR and RBC solvency
models . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.4.2 Risk management incentives in the SR and RBC mod-
els of the maximization of shareholders value in the
presence of FD costs . . . . . . . . . . . . . . . . . . 112
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

4 Enhancing insurer value via reinsurance and dividend dy-

namic optimization: multiple period models 126
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.2 Optimal reinsurance and dividend-payout strategies in the
maximization of shareholders value: a stochastic optimal con-
trol approach . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.2.1 Mathematical modelling of insurer surplus . . . . . . 132
4.2.2 Solutions to the particular stochastic control models
of insurers surplus . . . . . . . . . . . . . . . . . . . 138
4.2.3 The analysis of the value function under the optimal
control policy: risk-adjusted discount rate and the
companys level of capitalization . . . . . . . . . . . . 152
4.3 Reinsurance and dividend optimization in uncontrolled dividend-
barrier models of surplus . . . . . . . . . . . . . . . . . . . . 155
4.3.1 General results for Ito diffusion model . . . . . . . . 156
 ix

4.3.2 Analysis of the value function and the optimal divi-

dend barrier in Ito diffusion model of surplus . . . . . 161
4.3.3 Reinsurance optimization in an uncontrolled dividend
barrier model of surplus: an example of the model
governed by an arithmetic Brownian motion . . . . . 165
4.4 Reinsurance and dividend optimization in uncontrolled dividend-
barrier models of surplus with solvency restrictions . . . . . 166
4.4.1 Optimal dividend barrier policy and reinsurance under
solvency restrictions . . . . . . . . . . . . . . . . . . 168
4.4.2 Optimal dividend rate, leverage ratio and reinsurance
in the presence of insolvency cost: an expected utility
approach . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

5 Competitive equilibrium pricing in an integrated insurance-

reinsurance market 183
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.2 Comparative analysis of different models of insurance equilib-
rium pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
5.2.1 Economics of risk sharing in insurance-reinsurance mar-
ket . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
5.2.2 Financial equilibrium models of
insurance pricing . . . . . . . . . . . . . . . . . . . . 192
5.3 Walrasian equilibrium in integrated
insurance-reinsurance market . . . . . . . . . . . . . . . . . 197
5.3.1 Demand for insurance . . . . . . . . . . . . . . . . . 197
5.3.2 The insurers supply of insurance and demand for rein-
surance . . . . . . . . . . . . . . . . . . . . . . . . . 200
5.3.3 Supply of reinsurance . . . . . . . . . . . . . . . . . . 202
5.3.4 Dynamic equilibrium insurance and reinsurance pric-
ing. Numerical solutions . . . . . . . . . . . . . . . . 202
5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

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

3.1 Graphical illustration of excess of required minimal risk capital u1 (a, b)

under fixed level 10% of return on equity: exponential case . . . . . . 76
3.2 Graphical illustration of excess of required minimal risk capital u1 (a, b)
under fixed level 10% of return on equity: Pareto case . . . . . . . . 77
3.3 Graphical illustration of function V (a, b) defined on {a [0, 1]} {b
VaR [X]} . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.4 Graphical illustration of function u(a, b) defined on {a [0, a1 ]} {b
VaR [X]} . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
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]} . . . . . . . 84
3.6 Graphical illustration of the total return 1 + (umin , a, b) in model M1
with corporate tax = 30% . . . . . . . . . . . . . . . . . . . . . 96

x
 xi

3.7 Graphical illustration of the total return 1 + (umin , a, b) in model M1

with corporate tax = 15% . . . . . . . . . . . . . . . . . . . . . 98
3.8 Graphical illustration of the total return 1 + (a, b) in model M2 with
corporate tax = 30% . . . . . . . . . . . . . . . . . . . . . . . 99
Notations and Abbreviations

The following notations are frequently used in this dis-

sertation:

R the set of real numbers

E the operator of expectation
(, F, P) filtered probability space
a.s. almost surely
max A the maximum of the set A (discrete case)
sup A the maximum of the set A (continuous case)

The following abbreviations are frequently used in this

dissertation:

CAPM Capital Asset Pricing Model

CRRA Constant Relative Risk Aversion
FD Financial Distress
HJB Hamilton-Jacobi-Bellman Equation
IBNR Incurred But Not Reported
ODE Ordinary Differential Equation
OPT Option Pricing Theory
PDE Partial Differential Equation
PDMP Piecewise Deterministic Markov Process
RAROC Risk Adjusted Return on Capital

xii
 xiii

RBC Risk Based Capital

SDE Stochastic Differential Equation
SR Solvency Ratio
Chapter 1

Introduction

1.1 Reinsurance as an effective tool of in-

surer risk management in shareholders
value creation
Insurance companies are financial institutions that sell insurance contracts
to risk averse individuals (insureds or policyholders). Insurers collect premi-
ums from policyholders and are obliged to repay losses incurred from insur-
ance events. In other words insurers issue insurance debt through the sale
of insurance policies. Unlike debt of financial corporations, the insurance
debt is risky. This is because insurers experience considerable underwriting
risks due to loss severity and frequency of loss occurrence. There are two
important features that differentiate insurers from financial corporations:
their disadvantageous treatment of investments and their competitive ad-
vantages in raising insurance funds. On the one hand insurers operate in
much less beneficial tax and regulatory environment than financial compa-
nies, and they usually experience additional cost of double taxation. On
the other hand insurers have a competitive advantage in creating value by
borrowing in the insurance market since they have the expertise to manage

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:

avoid costly investment decision errors caused by under-investment;

alter their corporate capital structure so that tax payoffs are decreased;

decrease financial distress costs or insolvency costs by decreasing prob-

ability of insolvency.

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.

1.2 Motivation and structure of the thesis

The original idea of this thesis was to investigate the demand for reinsur-
ance in insurer risk management and to find out the impact of reinsurance,
as the classical risk transfer mechanism, on shareholder value. The starting
point was the investigation of the problem of finding the optimal functional
form of reinsurance under a mean-variance criterion. The key result in the
analysis of optimal reinsurance contract of endogenous form was developed
by Borch (1960 [14]) who considered the problem of minimizing the variance
of the total claims borne by the ceding insurer. Under fairly restrictive con-
ditions he proved that the stop-loss contract is optimal. However it should
Chapter 1 5

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

as the companys investors would prefer to measure the investment perfor-

mance of the supplied equity capital through the risk-adjusted discount rate,
under which the shareholders NPV of their investment project is zero. The
risk-adjusted discount rate is an effective measure of shareholders value cre-
ation. In Chapter Four of this thesis we explain the relation between the
two notions: expected presented value of future dividends and risk-adjusted
discount rate.
The idea of Chapter Five of this thesis is to review and compare differ-
ent models of insurance pricing, and to investigate the Walrasian equilibrium
insurance and reinsurance prices in an integrated insurance-reinsurance mar-
ket. The latter is motivated by Lin and Powers (see Lin and Powers (1999
[98])), who investigated a dynamic equilibrium model of risk-exchange be-
tween insureds and insurers in insurance market. Chapter Five extends
the Lin-Powers model by adding new agents - reinsurers, investigates the
equilibrium insurance and reinsurance prices, and analyzes the impact of
reinsurance market on the equilibrium insurance price.
The structure of this thesis is based on the chapters outlined above.

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

Chapter Three was presented at the Third Conference in Actuarial Sci-

ence & Finance on Samos (Greece) in September 2004, at the Third Actu-
arial Research @ UNSW Symposium in Sydney in November 2004.
Some results of Chapter Four were presented at the UNSW Actuarial
Research Seminar in Sydney in October 2004.
Chapter 2

Finding an optimal form of

reinsurance contract under
mean-variance criterion

In this chapter we consider the problems of finding an optimal reinsurance

contract endogenously (i.e. in a general functional form) under a mean-
variance criterion and for different insurance premium principles. As will be
shown below, under a mean-variance criterion the cedents expected profit
and variance of its portfolio after reinsurance are functionals dependent on
reinsurance coverage as a real function of losses. For many classical premium
principles these functionals are convex and Gateaux differentiable. These
properties allow us to apply the methods of convex non-linear programming
to find optimal reinsurance contracts under a mean-variance criterion. How-
ever, as it will be shown later in this chapter, the Esscher premium principle
is not convex, and for this case of premium principle the necessary conditions
of existence of an optimal reinsurance contract are studied.
Using convex programming the classical problem of finding an optimal
(re)insurance contract endogenously under the expected utility approach has
been reconsidered in this chapter for many classical actuarial principles that
are convex.
This chapter is organized in the following way. In Section 2.1 we overview

10
Chapter 2 11

early contributions to the research area of finding an optimal reinsurance

contract exogenously and endogenously under a mean-variance criterion. In
Section 2.2 we study problems of optimal reinsurance contracts for convex
premium principles. This section includes an introduction to convex anal-
ysis of functionals and applications in optimization. The particular cases
of convex premium principles such as mean, variance, standard deviation,
mixed variance-standard deviation, exponential are investigated. The nec-
essary conditions of existence of an optimal reinsurance under the Esscher
premium principle (non-convex premium principle) are determined.
Section 2.3 is devoted to the problem of finding an optimal (re)insurance
contract endogenously under the expected utility approach and generalized
convex premium principles.

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

Borchs problem from a different point of view. He restated and proved it

for a set of transformations wider than that which has been considered by
Borch. Further, Ohlin (1969 [111]) proved that under similar conditions,
stop-loss transformation retains its minimizing properties when the variance
is replaced by any member of a wide class of measures of dispersion. This
class of measures of dispersion has been defined as

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

problems of endogenous reinsurance optimization under mean-variance cri-

terion, or in other words the structure of the reinsurance premium directly
affects the cedents optimal decision of risk transfer from the cedent to rein-
surer.
An analogous situation is observed in insurance when we consider the
problem of optimal risk sharing between an insured and an insurer. Here
both the insured and the insurer maximize the expected utility of their
terminal wealths, and the degree of riskiness is characterized by their risk
aversion1 .
Optimal insurance decision studies can be divided into those in which
the form of insurance policy was exogenously specified (Gould (1969 [71]),
Arrow (1963 [5]), Mossin (1968 [107]), Smith(1968 [138])) and Blazenko
(1985 [13])), and those in which it was not (Borch (1960(a) [15]), (1960 [14],
Arrow (1971 [3]), Raviv (1979 [123]), Moffet (1979 [105]).
The results obtained by Moffet (1979 [105]), Raviv (1979 [123]), Spaeter
and Roger (1995 [140]), and many others such as Schlesinger (2000 [126]),
Gollier (2000 [69]) indicate 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 when the insurer is risk neutral and
the cost of insurance is proportional to the insurance indemnity, the optimal
insurance indemnity is the deductible policy (it is similar to the form of
stop-loss risk transfer in reinsurance). However, if 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).
1
In the case of reinsurance optimization under a mean-variance criterion we can ap-
proximate the mean-variance criterion by a corresponding quadratic utility function (see
Markowitz (1959 [100])) and then find optimal reinsurance under maximum expected
utility criterion.
Chapter 2 15

Going back to the problem of endogenous reinsurance optimization un-

der a mean-variance criterion, the expected profit net of reinsurance and
the variance of retained risk are just functionals of the risk transformation,
which is defined by a real measurable function from Hilbert space L2 . These
functionals are linear and directly defined by the reinsurance premium. Con-
sidering the reinsurance premium calculated under different premium princi-
ples such as variance, standard deviation, mixed variance-standard deviation
premium principles, exponential, Esscher (see Goovaerts et al (1984 [70])),
we notice that these premium principles are defined by linear functionals.
For all these premium principles their corresponding functionals are convex
(except Esscher) and Gateaux differentiable (see Deprez and Gerber (1985
[45])).
Hence, we conclude that the problem of endogenous reinsurance opti-
mization under a mean-variance criterion for most actuarial forms of rein-
surance premium is equivalent to a non-linear convex optimization problem.
In this chapter using standard methods of convex programming, we review
the problems of endogenous reinsurance optimization under mean-variance
criterion for different forms of reinsurance premium and derive the opti-
mal risk transfers. In the case of the Esscher reinsurance premium, which
is not convex, we provide necessary conditions of existence of the optimal
reinsurance strategy.
Finally, the problem of endogenous reinsurance optimization is consid-
ered under an expected utility criterion and the corresponding optimal form
of reinsurance contract is analyzed.
Chapter 2 16

2.2 Convex nonlinear programming methods

in finding an optimal reinsurance con-
tract
Throughout this section we will use the following notation. We consider Y as
a nonnegative random variable with continuous distribution function F (y)
defined on a given probability space (, F, P). An insurer is willing to buy
a reinsurance coverage to cover a random positive quantity R of Y , which
is a measurable positive function of Y , for a premium P . So the total claim
Y would split into a part paid by the reinsurer, R, and the one paid by the
e ), in such a way that Y = R(Y
insurer, say R(Y e ) + R(Y ). Therefore, we will

be solving reinsurance optimization problem on the set A of all admissible

reinsurance contracts that satisfy condition 0 R(Y ) Y . We also assume
that the premium paid by the insurer for the reinsurance arrangement is
calculated according to the following principle:

P[R(Y )] = E[R(Y ) + C (R(Y ))],

where C() is the risk loading represented by a bivariate function of R and

functional(s) dependent on reinsurance coverage R. As examples we consider
the following classical premium principles:
1) variance principle P[R(Y )] = E[R(Y )] + Var[R(Y )] with risk loading

C (R(Y )) = (R(Y ) E[R(Y )])2 , > 0;

p
2) standard deviation principle P[R(Y )] = E[R(Y )] + Var[R(Y )] with
risk loading

(R(Y ) E[R(Y )])2

C (R(Y )) = p , > 0;
Var[R(Y )]
Chapter 2 17

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

2.2.1 Reinsurance optimization for convex premium

principles

We consider the following reinsurance premium principle

P[R(Y )] = E[R(Y ) + C (R(Y ))],

Chapter 2 18

which is a functional defined on the set of admissible reinsurance contracts

A, and the reinsurers risk loading C is a bivariate function of R and func-
tional(s) dependent on R. We also assume that E[R2 (Y )] < , otherwise
in the case of generalized dispersion premium principle premium could not
be finite and in the case of the Esscher premium principle E[eR(Y ) ] = and
thus, the premium is undefined. Now, it is clear that R is a Hilbert space
p
with the norm kRk = E[R2 (Y )] for all R R, and A is a convex subset
of R. In addition, we assume that the functional P satisfies the following
invariance-under-translation property

P[R + c] = P[R] + c for all constants c R, (2.2.3)

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

Let F be linear continuous functional defined on Hilbert space R.

Definition 2.2.1. A functional F : R R is said to be convex on R, if for

all p1 > 0, p2 > 0 such that p1 + p2 = 1 and all R1 , R2 R

F[p1 R1 + p2 R2 ] p1 F[R1 ] + p2 F[R2 ]. (2.2.4)

Chapter 2 19

Clearly, if a functional F is convex on R it is also convex on any convex

subset of R.
As a simple example of convex functionals we can consider F[R(Y )] =
E[R(Y )], so that the mean value premium principle is convex.
It is possible to show that the generalized dispersion premium principle

P[R(Y )] = E[R(Y )] + (D[R(Y )]) ,

with positive, non-decreasing and convex function is also convex. Convex-

ity of this premium principle reduces to whether

(D[p1 R1 (Y ) + p2 R2 (Y )]) p1 (D[R1 (Y )]) + p2 (D[R2 (Y )]),

or, taking into account this fact that the real function is non-decreasing
and convex,

D[p1 R1 (Y ) + p2 R2 (Y )] p1 D[R1 (Y )] + p2 D[R2 (Y )].

The latter inequality holds and it follows from the following sequence of
rearrangements

D2 [p1 R1 (Y ) + p2 R2 (Y )] = p21 D2 [R1 (Y )] + p22 D2 [R2 (Y )]

+2p1 p2 Cov[R1 (Y )R2 (Y )] p21 D2 [R1 (Y )]

+p22 D2 [R2 (Y )] + 2p1 p2 D[R1 (Y )]D[R2 (Y )] = (p1 D[R1 (Y )] + p2 D[R2 (Y )])2 .

Thus the generalized dispersion premium principle is convex. From here it

follows that the standard deviation, variance and mixed standard deviation-
variance premium principles are convex (with the corresponding functions
(x) = x; x2 ; x + x2 , > 0, > 0).
It should be noticed that convexity is a general, and at the same time a
very strong property, because it implies the following useful properties.
Chapter 2 20

Lemma 2.2.1. (Superproportionality property) If a functional F :

R R is convex and satisfies invariance-under-translation property (2.2.3),
F[tR]
then t
is a non-decreasing function of t > 0.

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

which proves this lemma.

It should be noted that convexity does not imply a subadditivity property

of the functional F, i.e. F[R1 + R2 ] F[R1 ] + F[R2 ]. However, under
some additional condition on convex functionals, such functionals become
subadditive.

Lemma 2.2.2. (Proportionality property) A convex functional F that

satisfies (2.2.3) is subadditive if and only if it is proportional, that is F[tR] =
tF[R].

Proof. If F is proportional, then

1 1
F[R1 + R2 ] = 2F R1 + R2
2 2

1 1
2 F[R1 ] + F[R2 ] = F[R1 ] + F[R2 ] (2.2.6)
2 2

If a functional F is subadditive, then F[2R] 2F[R] for all R R. From

F[tR]
this and Lemma 2.2.2 it follows that t
must be constant in t.

We introduce another two properties which do not necessarily hold for

convex functionals. They are

F[R(Y )] max{R(Y )} (2.2.7)

Chapter 2 21

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.

Lemma 2.2.3. If a convex functional satisfies (2.2.7), then it satisfies also

(2.2.8).

Proof. Let R1 (Y ) R2 (Y ), then

p1
F[p1 R1 (Y )] = F p1 R2 (Y ) + p2 (R1 (Y ) R2 (Y )) p1 F [R2 (Y )]
p2

p1
+p2 F (R1 (Y ) R2 (Y )) p1 F [R2 (Y )] .
p2

By considering the limiting case when p1 1 we complete the proof of this

lemma.

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 ) ]

It satisfies property (2.2.7), however, as was shown by Gerber in (1981 [65])

(see also Deprez and Gerber (1985 [45])), property (2.2.8) does not hold in
general for the Esscher principle. This implies that the Esscher premium
principle is not convex.
There is also another criterion of convexity based on the convexity of a

real function h(t) = h(t; R, H) = F[R + tH], t [0, 1], R, H R.

Lemma 2.2.4. A functional F is convex on R if and only if the function

h(t) = F[R + tH] is convex for all R, H R.
Chapter 2 22

Proof. If F is convex, then

h(p1 t1 + p2 t2 ) = F[p1 (R + t1 H) + p2 (R + t2 H)]

p1 F[R + t1 H] + p2 F[R + t2 H] = p1 h(t1 ) + p2 h(t2 ).

If the function h(t; R, H) is convex for R, H R, then

F[p1 R + p2 H] = F[R + p2 (H R)] = h(p2 ; R, H R)

p1 h(0; R, H R) + p2 h(1; R, H R) = p1 F[R] + p2 F[H].

Corollary 2.2.5. A functional F is convex on the convex subset A R if

and only if the function

h(t) = h(t; R, H) = F[R + tH] = F[(1 t)R + tU ], t [0, 1]

is convex for all R, U, H = U R A.

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.

The second derivative of function h with respect to t is equal to

00 at + b
0
ac b2
00
hD (t) = at + 2bt + c =
2 = 3 0.
t at2 + 2bt + c t (at2 + 2bt + c) 2

So, again we reconfirm that the functional F is convex. Superposition of an

increasing function and convex functional F is convex, therefore general-
ized dispersion premium principle is a convex functional on A.
Chapter 2 23

Gateaux differentiability

Definition 2.2.2. A Gateaux derivative of a functional F : R R at

R R is any linear continuous functional OR F R such that

F[R + tH] F[R]

OR F[H] = lim
t0 t

for any H R.

We will say that a functional is Gateaux differentiable at R R if its

Gateaux derivative exists. As it is known from classical theory of functional
analysis, any Hilbert space is self-adjoint. In our case it means that R = R
and by the Riesz representation theorem for the functional OR F R there
exists a unique element G R such that

OR F[H] = E[G(Y )H(Y )] for all H R.

So, the element G uniquely determines the Gateaux derivative of functional

F and such that E[G(Y )] = 1. The latter equality follows from calculating
the Gateaux derivative at H(Y ) = c R\{0}:

F[R + tc] F[R]

c E[G(Y )] = OR F[c] = lim
t0 t
F[R] + tc F[R]
= lim =c
t0 t

On the other hand if the Gateaux derivative exists it can be represented by

d
dt
[h(t; R, H)]t=0 = h0 (0).
In this thesis we will use the function h for finding Gateaux derivatives.
Here we consider some useful examples of Gateaux differentiation.
1) Mean value: Gateaux derivative of functional E[] at R A is equal to

OR (E)[H] = h0 (0) = (E[R(Y ) + tH(Y )])0t=0

= (E[R(Y )] + tE[H(Y )])0t=0 = E[H(Y )],

Chapter 2 24

and its corresponding adjoint element is G(Y ) 1.

2) Generalized dispersion principle:

PD [R(Y )] = E[R(Y )] + (D[R(Y )])

Gateaux derivative of PD at R A is equal to

OR (PD )[H] = OR (E + (D))[H] = E[H(Y )] + ((hD (t)))0t=0

b
= E[H(Y )] + h0D (0)0 (hD (0)) = E[H(Y )] + 0 ( c)
c
Cov[R(Y ), H(Y )] 0
= E[H(Y )] + (D[R(Y )])
D[R(Y )]
and its corresponding adjoint element is equal to
R(Y ) E[R(Y )]] 0
G(Y ) = 1 + (D[R(Y )]).
D[R(Y )]
In particular, for the variance premium principle, when (x) = x2 , > 0,
and for the standard deviation principle, when (x) = x, > 0, Gateaux
derivatives of PD at R are equal respectively to

E[H(Y )] + 2Cov[R(Y ), H(Y )]

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.

Lemma 2.2.6. If for a functional F : R R there exists function h(t) =

h(t; R, H) = F[R + tH], t R, R, H R such that h00 (0; R, H) 0 for all
R, H R. Then F is convex on R.

Proof. For t 6= 0 we have

h (s; R + tH, tH) = F[R + (s + 1)tH] = h ((s + 1)t; R, H) , s R.

From here we obtain

0 h00 (0; R + tH, tH) = t2 h00 (t; R, H),

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

Hence, according to Lemma 2.2.6 the exponential principle is convex on R.

Yet another important property of convex and Gateaux differentiable
functionals is formulated in the following lemma.

Lemma 2.2.7. Let F : R R be convex and Gateaux differentiable func-

tional. Then the following inequality holds

F[U ] F[R] OR (F)[U R],

for all U, R R.

Proof. For any U, R R

F[U ]F[R] = h(1; R, U R)h(0; R, U R) h0 (0; R, U R) = OR (F)[U R]

since h is convex according to Lemma 2.2.4.

Non-linear convex reinsurance optimization

Consider a reflexive2 Banach space Z (note the Hilbert space R is a reflexive

Banach space) and let F0 , F1 be finite functionals on Z, and let S be a subset
of the space Z. We consider the problem

min F0 [Z],
ZS (2.2.9)
subject to F1 [Z] 0, Z S, .

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.

Theorem 2.2.8. (the Karush-Kuhn-Tucker theorem) Let the func-

tionals F0 , F1 and the underlying set S be convex. Then Z is a solution
to the problem (2.2.9) if and only if there exists Lagrange multiplier 0,
such that
1) F1 [Z ] = 0;
2) L [Z ] L [Z], Z S ,
where L [Z] = F0 [Z] + F1 [Z],

Remark 2.2.1. If the sufficient conditions for inequality constrained opti-

mality formulated in theorem 2.2.8 hold, then Z also minimizes F0 over
S= .
Indeed, according to the theorem 2.2.8 Z minimizes F0 over S and
moreover Z S= S

F0 [Z ] = L [Z ] L [Z] = F0 [Z].

Theorem 2.2.9. (the Karush-Kuhn-Tucker theorem (gradient form))

Let the functionals F0 , F1 be convex and Gateaux differentiable, and let the
set S be convex. Then Z is a solution to the problem (2.2.9) if and only if
there exists Lagrange multiplier 0, such that
1) F1 [Z ] = 0;
2) OZ (L )[Z] = 0, Z S .
Chapter 2 28

Proofs of these theorems can be found in Ioffe & Tikhomirov (1974 [86])
or Peressini et al. (1988 [118]).

2.2.2 Analysis of reinsurance optimization under mean

value, generalized dispersion and exponential rein-
surance premium principles

We start this subsection by considering the following classical mean-variance

model: an insurer (cedent) wants to buy reinsurance to reduce the variance
of its retained risk. So in this case the cedent would be interested to find an
optimal reinsurance contract that maximizes its expected profit after reinsur-
ance subject to the condition that the amount of the variance of the retained
risk does not exceed a predetermined level

max E[u + P P[R(Y )] (Y R(Y ))]
RA
subject to Var[Y R(Y )] constant,

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 underlying set A of all admissible reinsurance contracts is convex. It has

already been shown in the preceding subsection that the functional Var[Y
R(Y )] is convex and Gateaux differentiable. Now, if we assume that the
objective functional E[C(R(Y ))] is convex then we have to deal with a convex
optimization problem (2.2.10), which can be solved using Karush-Kuhn-
Tucker theorem.
Chapter 2 29

There is another convex optimization problem which is symmetrical to

the problem (2.2.10) and can be defined in the following way: minimize
variance of the retained risk subject to the condition that the expected value
of reinsurance coverage does not exceed a pre-specified level

min Var[Y R(Y )],
RA (2.2.11)
subject to E[C(R(Y ))] constant.

If in addition we assume that the functional E[C(R(Y ))] is Gateaux

differentiable and if the solutions to these two problems are determined
by positive Lagrange multipliers (theorem 2.2.9), then these solutions have
similar analytical (functional) forms.
Coming back to the famous research papers by Borch, Kanh and Ohlin,
we may conclude that the authors, solving the minimization problem (2.2.11)
under the mean value principle (i.e. C(x) = kx), showed that the stop-
loss reinsurance contract is the solution to this problem. This celebrated
result became classical in actuarial literature, however it has nowhere been
mentioned that this result is due to the special simple form of the premium
principle, and that under different convex premium principles this problem
has different solutions.
As a confirmation of this we may consider the results from the more
recent papers by Gajek & Zagrodny (2000 [59]) and Kaluszka (2001 [92]).
In fact the authors considered a convex optimization problem, similar to
(2.2.11), with constrained functional P[R(Y )] P defined by reinsurance
premium under generalized dispersion premium principle. This constrained
condition is equivalent to the constraint condition for the problem (2.2.11) if
we assume that C is considered under the mixed mean-generalized dispersion
premium principle, i.e.

E[C(R(Y ))] = 1 E[R(Y )] + 1 (D[R(Y )]) , 1 > 0, 1 > 0

Chapter 2 30

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.

Mean value premium principle

Let us first consider the classical method of risk theory of finding a particular
solution to the the problem (2.2.11).

Theorem 2.2.10. For every admissible reinsurance contract R A there

exists a unique retention r > 0 such that

E[R(Y )] = E[(Y r)+ ]

and
Var[Y R(Y )] Var[Y (Y r)+ ].

Proof. It is obvious that function E(r) = E[(Y r)+ ] is continuous decreas-

ing function of r [0, ) with values E(0) = EY and E() = 0. It implies
that there exists r > 0 such that E(r) = E[R(Y )].
Now, since E[Y R(Y )] = E[Y (Y r)+ ], the following expressions

Var[Y R(Y )] Var[Y (Y r)+ ]

(= Var[Y R(Y ) r] Var[Y (Y r)+ r])

Chapter 2 31

and
E[Y R(Y ) r]2 E[Y (Y r)+ r]2

have the same sign or equal 0.

Let us investigate the sign of the last expression. When y r then
|y R(y) r| 0 = |y (y r)+ r|. When y < r

y R(y) r y r = y (y r)+ r < 0.

It follows that |y R(y) r| |y (y r)+ r| and thus

Var[Y R(Y )] Var[Y (Y r)+ ].

In fact, we can analyze the result of theorem 2.2.10 in more detail. We

can plot the points of expected value and variance of the retained risk for
stop-loss contracts for different retentions r 0. The locus of all of such
points is an increasing, convex curve in the , -plane. First of all expected
value and variance of the retained risk are increasing functions of r:
d d
(r) = E[X (X r)+ ]
dr dr
Zr
d
= (1 F (x))dx = 1 F (r) > 0, r [0, ),
dr
0

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.

This implies that

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

following dual reinsurance optimization problem

min E[R(Y )]
RA

subject to Var[Y R(Y )] = constant.

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=

where L [R] = F0 [R] + F1 [R], > 0. The Gateaux derivative of the

functional L is equal to

OR (L )[H] = 2Cov[Y R(Y ), H(Y )] + E[H(Y )]

= 2E H(Y ) (Y R(Y ) + E[Y R(Y )]) = 0, H A= .
2

From here choosing H(y) = 2 (y R(y) + E[Y R(Y )] we conclude that

R (y) = y + E[Y R(Y )] = y d, d 0
2

Clearly d = 0 iff = 0 and R (y) y, which is impossible since E[R (Y )] =

l < E[Y ]. So d > 0 and taking into account 0 R (Y ) Y we conclude
Chapter 2 34

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.

From here we obtain

= (l) = 2d(l) 2E[Y (Y d(l))+ ]

d(l)
Z
= 2d(l) 2 xdF (x) + d(l) (1 F (d(l)))
0
Zd(l)
= 2d(l)F (d(l)) 2 xdF (x) > 0.
0

So, for (l) and R (Y ) = (Y d(l))+ we have

1) (l)F1 [R ] = (l) (E[R (Y )] l) = 0;
2) L [R] L [R ], R A= , since according to Lemma 2.2.7

L [R] L [R ] OR (L )[R R ]

= 2E[(R(Y ) R (Y ))(R (Y ) (Y d(l)))]

Zd(l)
= 2 R(y)(d(l) y)dF (y) 0.
0

Hence, according to the Karush-Kuhn-Tucker minimization criterion 2.2.8,

1) and 2) are sufficient conditions for R to be solution to the problem
(2.2.12).
In fact, according to Remark 2.2.1 to the Karush-Kuhn-Tucker theorem,
this optimal solution is the solution to the same problem with feasible region
A = {R A | F1 [R] 0}. The optimal solution to the problem (2.2.12)
is determined by the positive Lagrange multiplier, therefore the optimal
Chapter 2 35

solution to the symmetrical optimization problem

min E[R(Y )]
RA
Var[Y R(Y )] v < Var[Y ]
has the same analytical form, i.e. stop-loss reinsurance.

Generalized dispersion premium principle

Consider the generalized dispersion premium principle for a reinsurance con-

tract

PD [R(Y )] = E[R(Y )] + (D[R(Y )]),

p
where D[R(Y )] = Var[R(Y )] and () is continuous, non-decreasing and
convex function (one can notice that E[C(R(Y ))] = (D[R(Y )])). As was
mentioned earlier in this chapter this form of premium principle covers some
known premium principles, e.g. standard deviation, variance and mixed
standard deviation-variance.
Under this form of reinsurance premium principle the optimization prob-
lem (2.2.10) can be rewritten in the following way

min (D[R(Y )]),
RA (2.2.13)
subject to Var[Y R(Y )] v < Var[Y ].
Its symmetrical optimization problem is defined as follows

min Var[Y R(Y )],
RA (2.2.14)
subject to (D[R(Y )]) (l) < (D[Y ]).
Again, let us first consider the classical method of risk theory to deter-
mine a partial solution to this problem. Using the Cauchy-Schwartz inequal-
ity we obtain
p
D[R(Y )] = D2 [Y ] + D2 [Y R(Y )] 2Cov[Y, Y R(Y )]
p
D2 [Y ] + D2 [Y R(Y )] 2D[Y ]D[Y R(Y )] (2.2.15)

= D[Y ] D[Y R(Y )],

Chapter 2 36

and

D2 [Y R(Y )] = D2 [Y ] + D2 [R(Y )] 2Cov[Y, R(Y )]

D2 [Y ] + D2 [R(Y )] 2D[Y ]D[R(Y )] (2.2.16)

= (D[Y ] D[R(Y )])2 .

Here in general, the Cauchy-Schwartz inequality in a Hilbert space R with

p
the norm kRk = hR, Ri = E[R2 (Y )] becomes equality if and only if R(y) =
ay + b. Indeed, for T (Y ) = Y E[Y ], U1 (Y ) = Y R(Y ) E[Y R(Y )],
hUi ,T i
U2 (Y ) = R(Y ) E[R(Y )] and ai = hT,T i
, i = 1, 2 we have

0 kUi ai T k = hUi ai T, Ui ai T i = hUi ai T, Ui i

|hT, Ui i|2
= hUi , Ui i ai hT, Ui i = hUi , Ui i .
hT, T i

Thus, |hT, Ui i|2 hT, T ihUi , Ui i and kUi ai T k = 0 if and only if Ui ai T =

0.
If we assume R A to be continuous, then the optimal solution has the
form of quota share proportional reinsurance.
Using convex programming we can reveal all local solutions to problems
(2.2.13) and (2.2.14), and show that they belong to the class of change-loss
reinsurance contracts.
Let us reconsider this problem using the methods of convex analysis. In
this case F0 [R] = D[R(Y )] and F1 [R] = Var[Y R(Y )]v. Both functionals
F0 and F1 are convex and Gateaux differentiable. The feasible region for
this problem, A = {R A | F1 [R] 0}, is convex. Therefore, according to
the Karush-Kuhn-Tucker theorem (2.2.9) R is the solution to the problem
(2.2.13) iff it is the solution of the equation
1) F1 [R] = 0;
2) OR (L )[H] = 0, H A ,
where L [R] = F0 [R] + F1 [R], 0.
Chapter 2 37

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 )]

= (E[R(Y )H(Y )] E[R(Y )]E[H(Y )]) c

2 (E[(Y R(Y ))H(Y )] E[Y R(Y )]E[H(Y )])

= E [H(Y ) ((R(Y ) E[R(Y )]) c 2 (Y R(Y ) E[Y R(Y )]))]

2 2
= (2 + c ) E H(Y ) R(Y ) y+ E[Y ] E[R(Y )]
2 + c 2 + c
= 0,

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

where in general 0 a < 1 since 0 and c > 0. In fact a must be strictly

positive, otherwise R is constant which is not a solution. So the Lagrange
1 a
multiplier is also strictly positive and = 2 1a
c > 0. Now the following
forms of reinsurance

R0 (y) = ay, a (0, 1);

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 ]

= (2 + c ) E[(R(Y ) R (Y ))(R (Y ) ((ay + b)))]

Zdi
= (2 + c ) R(y)((ay + b))dF (y) 0.
0

Hence, according to the Karush-Kuhn-Tucker minimization criterion 2.2.8,

1) and 2) are sufficient conditions for Ri to be (local) solutions to the problem
(2.2.13).
If we are interested in continuous transformation of the risk, then the
optimal solution has the form R(y) = R0 (y) = ay, a [0, 1]. It follows
from constraint condition of problem (2.2.13) that (1 a)2 D[Y ] = v or

v
equivalently a = 1 (0, 1). From here we obtain the Lagrange
D[Y ]

1 a 0 (aD[Y
]) 0 (D[Y ] v )
multiplier = 2 1a c = 2(1a)D[Y ] =
2 v
> 0.
Since the Lagrange multiplier corresponding to the optimal solution
to the problem (2.2.13) is strictly positive, one of the local solutions to the
problem (2.2.14) has also the form of quota share proportional reinsurance.
Indeed the corresponding a and are
l 1a D[Y ] l
a= , =2 =2 0 > 0.
D[Y ] ac (l)
Similarly considering R1 , we can find an optimal change-loss reinsurance
contract by solving

|b|
Var Y a Y =v
a +

for b = b(a, v) when a is fixed, and maximizing

2 |b(a, v)|
a Var Y
a +

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.

E[C(R(Y ))] = 1 E[R(Y )] + 1 (D[R(Y )]) , 1 > 0, 1 > 0,

using the same methods of non-linear convex optimization, it is possible

to show that the optimal reinsurance contract is a change-loss reinsurance,
which is the mixture of a quota share proportional reinsurance and a de-
ductible reinsurance. This is consistent with the result obtained by Gajek
& Zagrodny (2000 [59]) and Kaluszka (2001 [92]), who considered a convex
optimization problem, similar to (2.2.11), but with a somewhat different con-
straint condition. They minimized the variance of retained risk under the
constrained condition that the reinsurance premium under the generalized
dispersion principle does not exceed a pre-specified level. In other words,
the constrained functional P[R(Y )] P is defined by reinsurance premium
under generalized dispersion premium principle. This constrained condition
is equivalent to the constraint condition for the problem (2.2.11) with

E[C(R(Y ))] = 1 E[R(Y )] + 1 (D[R(Y )]) , 1 > 0, 1 > 0.

Exponential premium principle

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

L [R] = E[CEX P (R(Y ))] + Var[Y R(Y )],

and the Gateaux derivative of this Lagrangian is equal to

OR (L )[H] = OR (E[CEX P ()])[H] + OR (Var[1 ])[H]

E H(Y )eR(Y )
= E[H(Y )] 2Cov[Y R(Y ), H(Y )]
E [eR(Y ) ]
" #
eR(Y ) E eR(Y ) 2E eR(Y ) (Y R(Y ) E[Y R(Y )])
= E H(Y ) .
E [eR(Y ) ]

Solving the equation OR (L )[H] = 0, H R we obtain

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).

such that for Rk (y) = ky A

E[CEX P (Rk (Y ))] < E[CEX P (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

2.3 Reinsurance optimization in the case of

non-convex premium principle
In this section we consider the reinsurance optimization problem (2.2.10)
under the Esscher reinsurance premium. As it was shown earlier in sub-
section 2.2.1 the Esscher premium principle is not convex. Therefore we
cannot apply the Karush-Kuhn-Tucker criterion. However, we can formu-
late a necessary condition for a local solution to a non-convex problem with
an equality constraint.

Theorem 2.3.1. Let the functionals F0 , F1 be Gateaux differentiable, and

let S= = {Z S | F1 [Z] = 0}. If Z is a local solution to the problem (2.2.9),
then there exists Lagrange multiplier R, such that

OZ (L )[Z] = 0, Z S= .

Let us consider the following non-convex problem with equality con-

straint

E[R(Y )eR(Y ) ]
min E[eR(Y ) ]
E[R(Y )] ,
RA (2.3.1)

subject to Var[Y R(Y )] = v.
Both the objective functional and the constraint functional are Gateaux
differentiable. So, we can apply theorem 2.3.1 in order to investigate the
necessary condition of existence of a locally optimal solution to the problem
(2.3.1). The optimal solution R = R A to the problem (2.3.1) satisfies
the following equation

OR (PESS E)[V ] + OR (D2 [1 ])[V ] = 0, 0 and for all V R,

where

OR (D2 [1 ])[V ] = OR (D2 [Y R(Y )])[V ] = OR (D2 [Y ] + D2 [R(Y )]

2E[(Y E[Y ])R(Y )])[V ] = 2Cov[R (Y ), V (Y )] 2E[(Y E[Y ])V (Y )]

= 2Cov[Y R (Y ), V (Y )].
Chapter 2 43

So,

OR (PESS )[V ] E[V ] 2E[(Y R (Y ) (E[Y ] E[R (Y )]))V (Y )] = 0.

Substituting the Gataeux derivative of PESS we obtain

"
eR (Y ) (1 + R (Y ))E eR (Y ) E R (Y )eR (Y )
E V (Y ) 1
E2 [eR (Y ) ]
2((Y R (Y ) (E[Y ] E[R (Y )])))] = 0, V R.

This implies that

eR (y) (1 + R (y))E eR (Y ) E R (Y )eR (Y )
1
E2 [eR (Y ) ]
+2((R (y) y + E[Y ] E[R (Y )]) = 0, y 0

or equivalently R should obey the following functional equation

1 R (y) 1
y= e (aR (y) + b) + R (y) + c , y 0 (2.3.2)
2 2
E[eR (Y ) ]E[R (Y )eR (Y ) ]
where a = E[
> 0, b = , c = E[Y ]E[R (Y )] >
eR (Y ) ] E2 [eR (Y ) ]

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
.

2.4 Imperfectly competitive equilibrium in a

duopoly insurer - reinsurer

2.4.1 Mean value premium principle

We assume 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; the reinsurer determines its premium according to the mean value
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[Y ] E[Y R(Y )] (1 + )E[R(Y )]

= P[Y ] E[Y ] E[R(Y )],

where P[Y ] is the insurance premium. If the insurer takes reinsurance in

order to reduce the variance of its insurance portfolio, it will choose an
optimal reinsurance that maximizes the profit under the same reduced level
Chapter 2 45

of variance, that is stop-loss reinsurance

R1 (Y ) = (Y r1 )+ = arg min E[R(Y )], s.t. Var[Y R(Y )] = v .
RA

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

The latter problem is equivalent to

e e
Y R2 (Y ) = arg min Var[Y R(Y )], s.t. E[R(Y )] = E[Y (Y r1 )+ ] .
e
RA

e ) = Y R(Y ). There exists r2 0, such that

where R(Y

E[(Y r2 )+ ] = E[Y (Y r1 )+ ].

e ), such that E[R(Y

Then, it follows from Theorem 2.2.10 that for all R(Y e )] =

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.

2.4.2 Standard deviation premium principle

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[Y ] E[Y R(Y )] PD [R(Y )]

p
= P[Y ] E[Y ] Var[R(Y )],
Chapter 2 47

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

However, it can be shown using convex analysis that there is no reinsurance

contract from the set of admissible contracts which would be a solution
to the latter convex optimization problem. However, if we assume that
the reinsurers choice of reinsurance contract is restricted to quota share
proportional reinsurance, then the contract R1 (Y ) = a1 Y will be reciprocally
optimal for both the cedent and the reinsurer.

2.4.3 Esscher premium principle

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

Under such a reinsurance premium principle the cedents expected value

after reinsurance equals

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.

Here, in this subsection we will investigate optimization problem (2.4.2)

for reinsurance contracts exogenously specified by the following class of
change-loss reinsurance contracts

A1 = {R A : R(y) = a(y b)+ , a [0, 1], b [0, ), y 0} ,

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.

Var[Y a(Y b)+ ] = E[Y a(Y b)+ ]2 E2 [Y a(Y b)+ ]

b b
Z Z Z
= x e dx + [(1 a)x + ab] e dx xex dx+
2 x 2 x

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

Solving the quadratic equation Var[Y a(Y b)+ ] v = 0 with respect to

a we obtain
p
b+1 (b + 1)2 (1 v)(2eb 1)
a=
2 eb
(b+1)2
under condition v 1 2eb 1
= f1 (b). Requiring a 1 we have
p
a1 (b + 1)2 (1 v)(2eb 1) 2 eb (b + 1) = 1 b eb 0.

Therefore,
p
b+1 (b + 1)2 (1 v)(2eb 1)
a = a(b, v) = 1.
2 eb

Solving this inequality with respect to v we obtain

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

we conclude that Dv = {(a, b) : a = a(b, v), f1 (b) v}.

Now
R
a (x b)ea(xb) ex dx
E R(Y )eR(Y ) b
E[R(Y )] = R
E [eR(Y ) ]
F (b) F (0) + eab e(a1)x dx
b
Z a
eb
(1a)2
a (x b)ex dx = eb
aeb = f3 (a, b),
1 eb + 1a
b

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

The variance of the ceded risk under the reinsurance contract R A1 is

equal to
2
Z Z

Var [a(Y b)+ ] = (y b)2 ey dy (y b)ey dy = a2 2eb e2b .
b b

Now, by requiring PESS [R(Y ); ] = p we first solve eb with respect to a

Chapter 2 51

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 = 0.3; 0.5 function is negative on its domain of definition and

1

thus, for every fixed value p 0, 1 the variance of ceded risk takes its
p
minimum at point a2 = 1+p
= a1 . It follows that b2 = 0 and the variance
of ceded risk is always smaller for any quota share proportional reinsurance
contract offered by cedent than for other reinsurance contracts from R1 with
the same reinsurance premium income.
Chapter 2 53

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.

2.5 Convex reinsurance optimization under

more general measures of risk
When we calculate the variance of the cedents retained risk we take into
account both positive and negative deviations from the mean of the retained
risk. These deviations have the same weight, however only the positive devi-
ations (large claim amounts) are harmful for the cedent, whereas small values
(negative deviations) may even be beneficial. Thus, in this case variance is
not an ideal (proper) measure of the retained risk. A more satisfactory mea-
sure of the harm caused by the retained risk Y R(Y ) can be obtained if
the square deviation of the retained risk is replaced by an increasing, convex
function h of Y R(Y ) E[Y R(Y )]. Now, instead of the variance the
harmfulness of a retained risk Y R(Y ) is measured by the expected harm

[R] = E [h (Y R(Y ) E[Y R(Y )])] .

As an example we can consider a non-symmetric risk measure

1 (R) = E (Y R(Y ) E[Y R(Y )])+ ,

which is determined by the harm function h(y) = y + = max{0, y}. Let us
reconsider the optimization problem (2.2.11) under the risk measure

min [R],
RA (2.5.1)
subject to E[C(R(Y ))] constant
Chapter 2 54

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 + ) .

The corresponding risk measures are

1) 1 [R] = E (Y R(Y ) E[Y R(Y )])+ ;
2
2) 2 [R] = E (Y R(Y ) E[Y R(Y )])+ .

Under the mixed mean-dispersion reinsurance premium principle (i.e.

E[C(R(Y ))] = E[R(Y )] + D[R(Y )]) the following two results hold

Proposition 2.5.1. The optimal reinsurance contract that minimizes the

risk measure 1 of retained risk has the following form


0, y m


R (y) = y m, m < y M



M m, y > M

where M > m > 0 are endogenously defined.

Chapter 2 55

Proposition 2.5.2. The optimal reinsurance contract that minimizes the

risk measure 2 of retained risk has the following form

0, y m
R (y) =
a (y m), y > m,

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.

2.6 Optimal reinsurance under utility approach

and convex premium principle
In the preceding sections we considered the problems of finding an optimal
reinsurance contract under which

the cedents expected profit net of reinsurance achieves its maximum

subject to the constraints on the retained risk; or
4
Such type of risk transfer is common in reinsurance practice (see Panjer et al (1998
[115]), p.72.)
Chapter 2 56

the cedents retained risk is minimized subject to the fixed level of the
cedents expected profit.

In general it can be shown that the maximization of the cedents expected

profit subject to level of the retained risk is equivalent to the unconstrained
maximization of expected utility of the cedents profit. Here the correspond-
ing utility function is directly defined by the form of risk measure. For
instance, it has been already mentioned above that in the case where the
cedents risk is measured by variance, i.e. using the mean-variance opti-
mization criterion, the corresponding utility function can be approximated
by a quadratic utility function. Later in Chapter 4 it will be shown that
the corresponding utility function can be well approximated by an isoelastic
utility when the companys risk is measured by ruin probability.
In this section we consider the following problem of finding an optimal
reinsurance contract under an expected utility approach and a convex rein-
surance premium principle. Let us consider a single period model in which
all insurance premiums P[X] are collected at the beginning and all insur-
ance claims X are paid at the end of the period. The insurers initial surplus
before taking reinsurance is w0 . At the beginning of the period the insurer
cedes R(X) to the reinsurer and pays for this a reinsurance premium which
is calculated according to a certain premium principle . It is assumed
that the insurer is risk averse with a smooth and concave utility function,
and wants to find such an optimal risk transfer R (X) that maximizes the
expected utility of its terminal wealth net of reinsurance, i.e.

R = arg max E [u (w X [R(X)] + R(X))] ,

R

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

of the utility function u we have that the functional

F[R] = E [u (w X [R(X)] + R(X))]

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 ]

= E [(V (X) OR [V (X)]) u0 (w X [R (X)] + R (X))] .

This is equivalent to

E [V (X) u0 (w X [R (X)] + R (X))]

O [V (X)] =
R .
E [u0 (w X [R (X)] + R (X))]

As we know OR [V (X)] is the functional in Hilbert space L2 and thus is

uniquely defined by the element GOR L2 which plays the role of gradient,
that is

OR [V (X)] = E [GOR (X) V (X)] .

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))]

It is possible to show that the latter necessary condition is sufficient too.

Indeed, consider

u (w X [R(X)] + R(X)) u (w X [R (X)] + R (X))

u0 (w X [R (X)] + R (X)) [[R (X)] [R(X)] (R (X) R(X))]

= E [u0 (w X [R (X)] + R (X))] GOR (X)

[[R (X)] [R(X)] (R (X) R(X))] ,

Chapter 2 58

and then take the expectation and use Lemma 2.2.7 (p. 26) to obtain

E [u (w X [R(X)] + R(X))] E [u (w X [R (X)] + R (X))]

u0 (w X [R (X)] + R (X)) [[R (X)] [R(X)] (R (X) R(X))]

= E [u0 (w X [R (X)] + R (X))]

[[R (X)] [R(X)] E [GOR (X) (R (X) R(X))]] 0.

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

However, for any other combinations of analytical forms of u, including

isoelastic power utility that corresponds to mean-ruin probability optimiza-
tion criterion, and the problem of solving (2.6.1) for R is not trivial.

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

different cases of reinsurance premium principles.

These results of insurer risk management using reinsurance lead to inves-
tigation of more advanced problems of insurer risk management including
maximization of shareholders value using a general form of risk transfer
(reinsurance) as an underwriting hedge. This is the objective of the next
two chapters.
Chapter 3

Enhancing insurer value via

reinsurance optimization:
insurance economics approach
in a single period model

In this chapter we investigate the problems of reinsurance demand in an

insurers value creation.
This chapter is organized as follows. In Section 3.1 we overview early
contributions to the research area of finding the optimal reinsurance contract
under the criterion of minimum ruin probability. In Section 3.2, in contrast
to most of the existing literature, we study problems of finding an optimal
change-loss reinsurance contracts (a wide class of reinsurance contracts that
includes proportional and excess of loss reinsurance) under a new criterion
of optimality based on ruin probability and value at risk in a discrete setting
(a single period model). In Section 3.3 we reconsider the problem of Section
3.2 in the presence of corporate tax. In Section 3.4 we investigate the same
optimization problem in the presence of frictional costs such as financial

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

[24]) as a dynamic stochastic control problem with the control variables

including proportional, non-proportional reinsurance and investment.
The optimization of reinsurance has recently been considered by Schmidli
(2001 [128], 2002 [129]) for proportional reinsurance, by Dickson and Wa-
ters (1996 [47]) and Hipp and Vogt (2003 [81]) for the compound Poisson
case of risk process and excess of loss (XL) reinsurance, and by Taksar and
Markussen (2003 [144]) for proportional reinsurance and surplus as a diffu-
sion process.
All these investigations usually focus on minimizing the ruin probability
only and do not take into account economic aspects (objectives) of an in-
surance business such as an insurers expected profit, return on shareholders
fund, etc. The unconstrained minimizations of ruin probability is unrealistic
from the point of view of the modern theory of integrated risk management
for an insurance company, since it focuses on risk minimization only without
any explicit regard to the companys economic value. There are also other
recent studies (e.g. see Taksar (2000 [143]) and references therein) that max-
imize the expectation of the discounted future dividends (companys value),
paid by an insurer to its shareholders allowing for reinsurance, but they do
not take into consideration frictional costs such as corporate tax and costs
of financial distress.
In this chapter, we study the demand for reinsurance in a single period
model in the presence of corporate tax and cost of financial distress. In order
to construct an objective function of an insurer we should first understand
the nature and economic aspects of an insurance business. In contrast to
industrial companies, insurers do not generally leverage themselves via cap-
ital markets. They collect insurance premiums (borrow money) by issuing
debt in the form of insurance policies that pay the policyholders (lenders)
compensation (financial benefits) if pre-specified events occur. To create and
Chapter 3 64

then issue insurance contracts, insurers rely on diversification and financial

markets. By pooling contracts that are not perfectly correlated, aggregate
losses become more predictable. While pooling reduces uncertainty, unex-
pected losses may still arise, potentially jeopardizing the insurers ability to
meet its obligations. On the other hand, unlike bondholders who can effec-
tively reduce their credit risk exposure by holding a well-diversified portfolio
of bonds with different issuers, policyholders generally cannot mitigate in-
surer default risk in any cost-efficient way. Therefore, policyholders usually
accumulate their credit exposure with insurers, the financial strength of
which is assessed by rating agencies and/or regulators. Insurers satisfy regu-
latory requirements on solvency/security by holding risk capital in addition
to operating capital including a component of premium income. We define
the premium, net of administrative expenses, as consisting of the expected
value of claim loss and risk loading (or risk premium).
According to Daykin et al (1996 (pp.156-157) [41]), Nakada et al (1999
[110]) and SOA Economic Capital Calculation and Allocation Subgroup
(2003 [139]) the sum of the risk premium and risk capital determines the
value of economic capital. Like industrial companies, insurers are financed
by their principals (shareholders)(see Brealey and Myers (1999 [21]), and
Culp (2002 [34])). Shareholders of insurance companies supply risk capital
(equity capital or surplus) that is invested on their behalf in financial
assets. In so doing, shareholders expect to earn a fair return on invested risk
capital. Insurers create shareholders value through investment in assets and
borrowing in the insurance market, rather than in capital markets. It should
be noticed here that in the presence of frictional costs such as corporate and
individual taxes (double taxation)1 it is more costly for insurers to create
1
The problem of double-taxation in insurance is not present in every country. For
instance, in Australia it is reduced due to the use of tax imputation system, according
to which individual shareholders who receive assessable dividends from a company are
entitled to a credit for the tax paid by the company on its income.
Chapter 3 65

value from investment in financial markets compared to direct investment

funds. However, insurers have a competitive advantage in creating value
by borrowing in the insurance market since they can directly manage 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.
We assume that one of the main insurers objectives is to maximize share-
holders value under solvency constraints imposed by a Regulatory Author-
ity. An insurer may traditionally improve its solvency level or reduce in-
solvency risk, which captures both undesirable large fluctuations and ruin
probability (see Gerber (1979 [64]), and Hurlimann (1993 [84])), by buying
reinsurance to reduce the unexpected fluctuations in the insurance losses
of the insurer. In a regulated insurance market when the solvency level,
required by a Regulatory Authority, is fixed the insurer can maintain this
level by two control variables: reinsurance and risk capital. In fact there are
at least two possibilities to do this: 1) (model M1) risk capital is indepen-
dent of the future possible buying of reinsurance, and is at least the required
minimum value of the risk capital determined at the beginning of the period
without reinsurance; 2) (model M2) the required minimum value of the risk
capital is determined taking into account future purchasing of reinsurance.
While the first possibility gives two independent control variables, the
second one leaves only one control variable, reinsurance, since the required
capital is explicitly dependent of the ceding amount of insurance, and thus
can be expressed through the reinsurance control variable. Under the sec-
ond possibility, purchasing reinsurance will normally decrease required risk
capital, and decreasing risk capital will increase the demand for reinsurance.
By considering shareholders of the insurance company as residual claimants
it is natural to consider the measure of performance of risk capital defined
Chapter 3 66

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.

3.2 Demand for reinsurance in shareholders

value creation: one-period frictionless model
Consider a single period model. Let X denote the random aggregate claims
of an insurer portfolio PX and let X have the distribution function F (x), x
0 defined on the probability space (, P). It is assumed that in perfectly com-
petitive insurance and reinsurance markets, insurers are subject to the risk
of insolvency, however to simplify the analysis we assume reinsurers are not
subject to the risk of insolvency3 . At the beginning of the period an insurer
should satisfy solvency conditions required by a Regulatory Authority, i.e.
an insurer should hold an amount of capital (risk capital), in addition to the
2
In this dissertation by introducing corporate tax we consider frictional costs caused
by convexity of tax payment. We will show that risk hedging can reduce expected taxes
(or create value).
3
For instance, in some states of the European Union no solvency requirements are
imposed on reinsurers.
Chapter 3 67

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 :

VaR (X) = inf{x R|P[X > x] 1 }

The total premiums collected at the beginning of the period is equal to

P = (1 + )E[X], where > 0 is the insurers risk loading. Note P does
not depend on capital or reinsurance, i.e. it is not adjusted with respect
to the value of insolvency exchange option. It is assumed that there are
no investment earnings. Then the risk capital required by the Regulatory
Authority is an amount of capital u such that

P[{u + P X > 0}] .

Therefore,

u umin = VaR [X] (1 + )E[X].

Without reinsurance the return on the risk capital provided by share-

holders is equal to

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

there is a demand for reinsurance in maximizing the return on risk capital.

We consider the class of change-loss reinsurance contracts

J = {Ja,b ()|Ja,b (X) = a(X b)+ , a [0, 1], b [0, )} .

This class of exogenously pre-specified reinsurance contracts includes ordi-

nary quota share (proportional) and stop-loss (or excess of loss) reinsurance.
If a = 1 we have stop-loss reinsurance, and if b = 0 we have quota share
proportional reinsurance. We will investigate the demand for change-loss
reinsurance in the following two models:

M1) two control variables (risk capital and reinsurance):

E max{0,u+P (a,b)(XJa,b (X))}
maximize 1 + (u; a, b) = u
,
(3.2.1)
subject to u umin and (a, b) [0, 1] [0, ),

M2) one control variable (reinsurance):

E max{0,umin (a,b)+P (a,b)(XJa,b (X))}
maximize 1 + (a, b) = umin (a,b)
,
(3.2.2)
subject to (a, b) [0, 1] [0, ),

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).

3.2.1 Maximizing return on risk capital by reinsur-

ance and risk capital

Consider model M1 of reinsurance-risk capital optimization

max E max{0,u+P (a,b)Ia,b (X)} ,
u, a, b u
(3.2.3)
u umin , (a, b) [0, 1] [0, ),

where umin = VaR [X] (1 + )E[X]. After taking reinsurance from class
J the cedents premium income becomes

P (a, b) = P (1 + )E[Ja,b (X)] = (1 + )E[X] (1 + )aE[(X b)+ ],

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

P = (1 + )EF [X] = EG [X],

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

From (3.2.4) we obtain

R
(1 F (x))dx
1 bk 1
1 + (b, ) = R > = 1 + .
k k
(1 F (x))dx
b

It is obvious that the premium income P (a, b) is positive for all a [0, 1]
and b [0, ], indeed

P (a, b) = (1 + )E[X] (1 + (b, ))aE(X b)+

Z

= (1 + ) E[X] a (1 F (x))dx

> 0. (3.2.5)
b
1+

The cedents after-reinsurance surplus is equal to

S(umin + u1 ; a, b) = umin + u1 + P (a, b) Ia,b (X))

= VaR [X] P + u1 + P (a, b) Ia,b (X),

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

(umin + u1 ; a , b ) = max (umin + u1 ; a, b).

u1 [0,), a[0,1], b[0,)
Chapter 3 71

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.

FIa,b (y) = P[X a(X b)+ y] = P[{X a(X b)+ y} ]

= P[{X a(X b)+ y} {X > b}]

+P[{X a(X b)+ y} {X b}

= P[{(1 a)X + ab y} {X > b}] + P[{X y} {X b}]

y ab
= P X {X > b} + P[X min{y, b}]
1a

y ab
= 1{b< yab } P b < X + F (min{y, b})
1a 1a

y ab
= 1{b< yab } F F (b) + F (min{y, b})
1a 1a
(
F (y), y < b;
= ,
F yab
1a
, y b.

where 1A is indicator of an event A. As we can see under a < 1 the distribu-

tion function FIa,b is continuous. In the trivial case, when a = 1 (stop-loss
reinsurance) we have
(
F (y), y < b;
FI1,b (y) = 1{b<y} (1 F (b)) + F (min{y, b}) =
1, y b.

Theorem 3.2.1. There is no demand for change-loss reinsurance in the

model M1, and moreover


{u = umin ; a = 0 b = } = arg max (u; a, b) .
u[umin ,), a[0,1], b[0,)
Chapter 3 72

Proof. Let us calculate the cedents expected terminal value.

E [max {0, S(umin + u1 ; a, b)}]

= E [max {0, VaR [X] P + u1 + P (a, b) Ia,b (X)}]

Z
= E max 0, VaR [X] + u1 (1 + (b, ))a (1 F (x))dx Ia,b (X)


b
(uZ1 , a, b)

= ((u1 , a, b) y)dFIa,b (y), (3.2.6)

0

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

So, for a < 1

(uZ1 , a, b)

((u1 , a, b) y)dFIa,b (y)

0
(uZ1 , a, b) (uZ1 , a, b)

= FIa,b (y)dy = 1{(u1 , a, b)<b} F (y)dy

0 0

Zb (uZ1 , a, b)
y ab
+1{(u1 , a, b)b} F (y)dy + F dy
1a
0 b
Chapter 3 73

(uZ1 , a, b)

= 1{(u1 , a, b)<b} F (y)dy

0
(u1 , a, b)ab

Zb Z
1a

+1{(u1 , a, b)b}
F (y)dy + (1 a) F (y) dy
.
0 b

For every fixed u1 and b we have

(uZ1 , a, b) Z

F (y)dy = (1 + (b, ))F ((u1 , a, b)) (1 F (y))dy < 0
a
0 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)

lim ((u1 , a, b) y)dFIa,b (y) = ((u1 , 1, b) y)dFI1,b (y).

a1
0 0

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)

((u1 , a, b) y)dFIa,b (y) = ((u1 , a, b) y)dF (y)

0 0
(uZ1 , a, b) (uZ1 , 1, b)

= F (y)dy F (y)dy, when a 1.

0 0
Chapter 3 75

2) If for some fixed b (u1 , 1, b) b then a (u1 , a, b) b and

(uZ1 , a, b)

(u1 , a, b)ab


Z b Z1a

lim FIa,b (y)dy = lim F (y)dy + (1 a) F (y)dy
a1 a1



0 0 b


RK


Zb F (y)dy


b
= lim F (y)dy + 1 ,
a1

1a

0

1
R
where K = 1a
VaR [X] + u1 a (1 + (b, )) (1 F (x))dx + b . The
b
latter limit can be found using LHopitals rule.
So,
b
(uZ1 , a, b)
Z (u1 , 1, b) b F
(u1 , a, b)ab
1a

lim FIa,b (y)dy = lim F (y)dy + 1
a1 a1 (1 a)2
(1a)2
0 0
Zb Zb
= F (y)dy + (u1 , 1, b) b = ((u1 , 1, b) y)dF (y)
0 0
(uZ1 , 1, b)

+((u1 , 1, b) b)(1 F (b)) = ((u1 , 1, b) y)dFI1,b (y).

0

Therefore, the function (, a, ) decreases on [0, 1], and there is no demand

for reinsurance.
Finally
VaRR[X]+u1
F (y)dy
0
(umin + u1 , 0, ) =
u1 u1 VaR [X] + u1 (1 + )E[X]
VaRR[X]+u1
F (VaR [X] + u1 ) (VaR [X] + u1 (1 + )E[X]) F (y)dy
0
=
(VaR [X] + u1 (1 + )E[X])2
!
VaRR[X]+u1 R
F (VaR [X] + u1 ) (1 F (y))dy (1 + ) (1 F (y))dy
0 0
< 0,
(VaR [X] + u1 E[X])2
Chapter 3 76

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 = ).

Even for simple forms of distribution function F it is impossible to ex-

press u1 as a simple function of a and b. However we can use some numerical
examples to explicitly plot the excess u1 (a, b) of the required minimal risk
capital. Here we consider two examples:
1) aggregate loss X is exponentially distributed with cdf F (x) = 1 e0.01x
(light tail distribution);
100
2
2) aggregate loss X is Pareto distributed with cdf F (x) = 1 100+x
(heavy
tail distribution);
For both distributions the mean is equal to 100. It is assumed for these
examples that the required solvency level is = 97.5%, the insurers risk
loading = 0.4 (i.e. risk adjustment coefficient k = 0.7143). In Figures
Chapter 3 77

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.

3.2.2 Maximizing return on risk capital by reinsur-

ance

Consider model M2 of reinsurance optimization

max E max{0,umin (a,b)+P (a,b)(XJa,b (X))} ,
a, b umin (a,b)
(3.2.7)
(a, b) [0, 1] [0, ),

with required minimal risk capital

umin (a, b) = VaR [X Ja,b (X)] P (a, b),

Chapter 3 78

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

S(a, b) = umin (a, b) + P (a, b) (X Ja,b (X))

= VaR [Ia,b (X)] Ia,b (X).

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

= P [Ia,b (X) < VaR [Ia,b (X)]]

F (VaR [Ia,b (X)]), VaR [Ia,b (X)] < b;
=
F VaR [Ia,b (X)]ab , VaR [Ia,b (X)] b,
1a

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.

and by Definition 3.2.1

(
VaR [X], VaR [X] < b;
VaR [I1,b (X)] =
b, VaR [X] b.

Summarizing, we conclude that a [0, 1]:

(
ab + (1 a)VaR [X], b VaR [X];
VaR [Ia,b (X)] =
VaR [X], b > VaR [X].
Chapter 3 79

The required minimal risk capital under change-loss reinsurance is equal to

u(a, b) =
R


ab + (1 a)VaR [X] + (1 + (b, ))a (1 F (x))dx (1 + )E[X], b VaR [X];

b
= R
(3.2.8)


VaR [X] + (1 + (b, ))a (1 F (x))dx (1 + )E[X], b > VaR [X].
b

The cedents terminal value is equal to

(
ab + (1 a)VaR [X] Ia,b (X), b VaR [X];
S(a, b) =
VaR [X] Ia,b (X), b > VaR [X],
and the expected value of its limited liability at the end of the period is

V (a, b) = E[max{0, S(a, b)}] (3.2.9)

Z
(a,b) Z
(a,b)

= ((a, b) y) dFIa,b (y) = FIa,b (y)dy,

0 0

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

In the case where a = 1 I1,b (X) is a stop-loss transformation of the loss X,

and thus the cdf FI1,b (y) of transformed loss has a jump at point y = b. For
b VaR [X]: (a, b) = b and according to (3.2.9)
Zb
V (a, b) = (b y)dFI1,b (y)
0
Zb Zb
= (b y)dF (y) + (b b)(1 FI1,b (b)) = F (y)dy,
0 0
Chapter 3 80

400
300
VHa,bL
200
300
100
0 200
0
b
0.25
0.5 100
a 0.75
10

Figure 3.3: Graphical illustration of function V (a, b) defined on {a [0, 1]} {b

VaR [X]}

for b > VaR [X]: (a, b) = VaR [X] and V (a, b) =

Z [X]
VaR Z [X]
VaR

= (VaR [X] y)dFI1,b (y) = (VaR [X] y)dF (y)

0 0
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+

thus u(a, ) decreases on (VaR [X], ).

Moreover,

lim u(a, b) = (1 a)(VaR [X] P ) and

b0
(3.2.13)
lim u(a, b) = (VaR [X] P ).
b

It follows from (3.2.11)-(3.2.13) that the global minimum of u(a, b) is at-

tained at a = 1 and b = 0 (full reinsurance) and equal to 0. In other words
the purchase of full reinsurance reduces the required risk capital to 0. But
Chapter 3 82

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 () =

Using (3.2.5) we obtain

1 + 1 + (b, ) b 1 + b

P (a, b) = ae = 1 ae 1+ . (3.2.14)

The cedents terminal value defined in (3.2.10) becomes

V (a, b) =

a b + ln(1) + 1 eb (1 ) ln(1)
, b [0, VaR [X]];

=
ln(1)
, b (VaR [X], ).

The required risk capital on {a [0, a1 ]} {b [0, VaR [X]]} defined in

Chapter 3 83

300

uHa,bL 200
100 300

0 200
0
b
0.25
0.5 100

a 0.75
10

Figure 3.4: Graphical illustration of function u(a, b) defined on {a [0, a1 ]} {b

VaR [X]}

(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)

on the interval b [0, VaR [X]].

Chapter 3 85

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

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)
is
(
A 1 L 1 , A1 > L 1
E1 =
0, A1 L1 ,

and the terminal value of insurers liability is

(
L1 , A1 > L1
1 =
A1 , A1 L1 .

So, the fair insurance premium is

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 } .

Summarizing, we conclude that the equilibrium insurance premium is a

solution to the following system of two equations

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

By introducing a change-loss reinsurance we transform the companys

liability from 1 to
(
b1 = L1 a(L1 b)+ , A
L b1 > L
b1
b1 =

b1 , A
A b1 L
b1 ,
h i
the premium P0 to Pb0 = er EQ
b 1 and solve the same system with respect

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.

3.3 Demand for reinsurance in shareholders

value creation: one-period model in the
presence of corporate tax
In this section the problem of demand for change-loss reinsurance in a single-
period model in the presence of a corporate tax is studied.
As was shown in the previous section the decision to reinsure is an im-
portant tool of altering a companys capital structure, which in turn gives
an opportunity to create (enhance) shareholders value. However, in the
paper by Garven (1987 [61]) the author suggests that in order for insurer
capital-structure decisions (including reinsurance) to matter in any meaning-
ful sense, factors such as frictional capital costs, including tax shields, agency
and financial distress costs must be considered. Indeed, unlike investment
Chapter 3 88

funds, insurers may be subject to additional corporate tax and operate in

a highly regulated environment where regulations are designed to protect
policyholders. These frictions generate a need to provide shareholders with
an additional return on the risk capital they supply over and above the base
cost of capital. There are essentially three sources of frictional capital costs:
costs of double taxation, costs of financial distress and agency costs.
In this section we will consider costs of double taxation only. Insurance
companies are taxed on their earnings: investment return and underwriting
profit before it is distributed to shareholders. This generates an additional
cost component relative to an investment fund. In practice the tax sched-
ules facing firms typically are convex i.e., higher levels of corporate earnings
usually encounter higher rates of marginal taxation. Even in the case of flat
rate of corporate tax the valuable tax payment has a convex form (form of
max{0, } function). Given tax convexity, Jensens inequality implies that
expected taxes will be reduced if the riskiness of earnings is reduced. If
earnings are quite volatile (risky), upside variation causes a large increase
in taxes but downside variations causes a little reduction of taxes. There-
fore, earnings stabilization will avoid the large potential upside increase in
taxes without sacrificing much of a tax decrease on the downside. Insurers
may control their underwriting risk (stabilize their earnings) by purchasing
reinsurance to create value by reducing expected taxes.
We assume that the cost of taxes arising out of the insurance transac-
tions and insurers investment should be included in the risk loading (risk
premium) paid by policyholders. The reason is as follows: when writing a
policy the insurer commits equity capital (risk capital required by the reg-
ulator) to the insurance business. The owners (principals) of the insurance
company always have the alternative of not writing insurance and invest-
ing their capital directly in financial assets (shares and bonds) in a capital
Chapter 3 89

market. They will not enter into an insurance transaction if by doing so

they subject income on their investment to an additional layer of taxation.
Therefore, the policyholders must pay the tax to provide a fair after-tax re-
turn on equity capital. According to the set up of the Myers-Cohn model of
determining the fair insurance premium (see Myers and Cohn (1987 [109])),
the premium is defined as fair if the insurer is indifferent between selling the
policy and not selling it. The insurer will be indifferent if the market value
of the insurers equity is not changed by writing the policy.
We reconsider models M1 and M2 of the reinsurance optimization by
taking into account
1) the possibility to reinvest both risk capital (equity capital) and premium
income at the beginning of the period;
2) corporate tax on underwriting profits and investment income
The return on investment i is a random variable. Claim costs are as-
sumed to be independent of return on investment. Underwriting profits and
investment income are taxed at the end-of period, at the rate , if taxable
earnings are positive, and the residual funds are distributed to shareholders6 .
So, again, using an explicit formula we can define the cedents expected
value of after-tax limited liability. Before tax, the shareholders have a valu-
able claim, if the terminal value of cash flows derived from the cedents un-
derwriting, reinsurance and investment activities is non-negative only. The
expected value of this valuable claim
6
In a multi-period model, if taxable earnings are negative but the direct insurer is
still solvent, then it receives a tax shield equal q , where q 1. In other words, an
insurer carries taxable (at rate q ) losses forward to offset future income. If the insurer is
insolvent, then the tax shield on losses is equal to zero. Here in this section we consider
a single period model, and thus assume that q = 0
Chapter 3 90

within model M1 is equal to

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

within model M2 is equal to

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

1 = (1 + i) = (1 + i) (u(a, b) + P (a, b)) = (1 + i)VaR [Ia,b (X)]

(
(1 + i) (ab + (1 a)VaR [X]) , b VaR [X];
=
(1 + i)VaR [X], b > VaR [X].

If the taxable total amount of investment income and underwriting profit

after reinsurance is non-negative, then the government will have a valuable
claim. The expected value of this taxable amount is equal to:

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

within model M2 this taxable amount is equal to

VeT (a, b) = Ei EIa,b (X) [max {i (u(a, b) + P (a, b)) + P (a, b)

Z2
Ia,b (X); 0} | i ]] = Ei (2 y) dFIa,b (y) ,
0

where

2 = iu(a, b) + (1 + i)P (a, b) = 1 u(a, b)

= iVaR [Ia,b (X)] + P (a, b).

The total shareholders expected after-tax terminal value is equal to:

within model M1

V (u, a, b) = VS (u, a, b) VT (u, a, b) (3.3.1)

Z1 Z2
= Ei (1 y) dFIa,b (y) (2 y) dFIa,b (y) ,
0 0

within model M2 it is equal to

Ve (a, b) = VeS (a, b) VeT (a, b) (3.3.2)

Z1 Z2
= Ei (1 y) dFIa,b (y) (2 y) dFIa,b (y) .
0 0

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

We should notice that, in fact,

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+

The following equality also holds for b {2 > b}

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];

and 2 = 1 u(a, b) = i + P (a, b) =

(
g3 (a, b; , i), if b VaR [X];
=
g4 (a, b; , i), 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 ,

2) b D {b > (1 + i)VaR [X]} (that is b > 1 > 2 )

1 1
1 2
1
a,b (i) = 1 1e 2 1e ,
u(a, b)

where 1 = (1+i) ln(1)

and 2 = i ln(1)

+ 1 (1 + ) a(1 + (b, ))eb .
We can get an analogous formulae to calculate u,a,b (i) in the model M1.

1 b
1 a b ab
11a
u,a,b (i) = 1 1e e e 1{1 b}


1 1

+ 1 1e 1{1 <b}


1 b
1 a b ab
21a
2 1e e e 1{2 b}


1 2
1
+ 2 1e 1{2 <b} ,
u
Chapter 3 96

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

The following two figures represents graph 1 + (umin , 1, b) on intervals b

[0, VaR [X]] and b [VaR [X], ).
1 + 1 +

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%

It seems that there is no demand for reinsurance in model M1 for low

values of corporate tax (e.g. see the Table 3 and also the graph of the total
7
The figures in the table are high from practical point of view. This table serves as an
illustration example only.
Chapter 3 98

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%

return on equity in model M1 with corporate tax = 15% in the Figure

3.7). This is somewhat expected, since it was shown in the previous section
that model M1 does not induce demand for reinsurance in maximization of
return on equity in a frictionless environment. So including only a small
amount of frictional costs (e.g. corporate tax) in model M1 may not affect
the optimal reinsurance strategy.
In model M2 we observe demand for change-loss reinsurance (see Figure
3.8). The following two figures represent the graph 1 + (a, b) on intervals
b [0, VaR [X]] and b [VaR [X], ).
1 + 1 +

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.

3.4 Demand for reinsurance in shareholders

value creation: one-period model in the
presence of costs of financial distress
One of the main incentives to manage corporate risk assumed by a corpora-
tion, is that variability in financial outcomes is costly when the corporation
operates in an environment with frictional costs. The risk management
strategies depend on the nature of the frictional costs. There are essentially
three sources of frictional capital costs: tax asymmetry, costs of financial
distress and agency costs. In the insurance industry, where the market is
regulated, an insurer, which is effectively leveraged by risky insurance debt,
experiences an additional two costs of corporate risk: cost of double taxation
and cost of regulatory restrictions. This section investigates the strategies of
an insurance firms risk management in the maximization of shareholders
value under the presence of financial distress costs.
Briefly financial distress can be defined as a low net-worth (surplus) state
of the insurance company in which the insurer incurs additional deadweight
losses. The notion that financial distress is a different state from legal in-
solvency has been introduced in the financial literature (e.g. see paper by
Diamond (1991 [46]), where the model of an illiquid but solvent firm is con-
sidered, and the paper by Froot et al. (1993 [58]), where the model of a
firm with low cash-flow is considered; also see the paper by Briys and de
Chapter 3 101

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

companys surplus net of regulatory surplus and FD costs, which depends

on the terminal value of the companys surplus as well as on the path of the
companys surplus during the period.
In this section we consider four different forms of financial distress costs.
These are

1) proportional to the terminal value of the insurers assets;

2) proportional to the terminal value of the insurers surplus (assets minus

liabilities);

3) constant;

4) reduced upside potential of the terminal value of insurers surplus.

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

We investigate the second model of the maximization of shareholders value

under the presence of of the other forms of FD costs.

3.4.1 Models setup and definitions: SR and RBC sol-

vency models

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

dLt = L dt + L dWtL , t [0, T ], (3.4.1)

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

dt = dt + dWt , t [0, T ], (3.4.2)

Chapter 3 105

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 distressed and solvent;

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

Within the SR model the expected present value of shareholders terminal

payoff is equal to

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:

using reinsurance to control underwriting risk L , or

using both reinsurance and optimal investment strategies to control

the volatility of the solvency ratio.

The RBC model

In this subsection we consider the insurers underwriting process over the

period of time [0, T ] between two consecutive regulatorys audits and define
insurers solvency within the RBC model in terms of risk based capital and
dynamics of the companys surplus S as a stochastic process St , t 0. Let
S S0 denote the minimal capitalization level (regulatory surplus or risk
based capital) at which the insurer is considered financially solvent by an
insurance regulator, and the insurer will be announced insolvent at time
= inf{t {T, 2T, ...} : St < S }. It is assumed that at time the
insurers surplus will be fairly divided among its creditors, so its surplus will
Chapter 3 107

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

surplus net of regulatory capital and FD costs:

V0RBC = erT EQ (ST S )+ 1{FinHealthy} (3.4.4)

+(F (ST ) S )+ 1{FinDistressed & Solvent}
h i
rT + +
= e EQ (ST S ) 1{mST >D} + (F (ST ) S ) 1{mST D} ,

where Q is an equivalent martingale measure, that exists under the assump-

tion that the market for assets and liabilities is arbitrage free, but incomplete
(as it is typical in insurance).
The value V0RBC in (3.4.4) can be rewritten in the following way
h i
V0RBC = erT EQ (ST S )+ 1{mST >D} + (F (ST ) S )+ 1{mST D}
h i
= erT EQ (ST S )+ (ST S )+ (F (ST ) S )+ 1{mST D}
h
= erT EQ (ST S ) (ST F (ST )) 1{mST D}{F (ST )>S }
i

+ (S ST ) 1{ST S } + 1{mST D}{ST >S F (ST )} (3.4.5)

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

An insurer can use two possibilities to control investment risk:

1) it can construct an investment portfolio with optimal volatility; or
2) the volatility can be fixed and an insurer can reduce risk by buying deriva-
tive contracts such as options.
The underwriting risk can be reduced by traditional cession (risk trans-
fer), i.e. through purchase of reinsurance contracts in the reinsurance mar-
ket. It is worth noticing that to some extent the reinsurance contract re-
sembles derivative contracts for financial risk. Using both, an agent can
reduce risk (volatility of investment portfolio or underwriting risk, which is
traditionally measured by the probability of insolvency).
So far we have considered the insurers surplus in the general form of
a stochastic process. To be able to calculate the value V0RBC in (3.4.5) we
have to be more specific about the model of the companys surplus. The
following defines the dynamics of the companys surplus under the physical
probability measure P.
Dynamic model of the insurance companys surplus. We con-
sider the diffusion model of an insurers surplus formulated by Powers in
(1995 [121]). In practice insurance firms receive insurance premiums and
must establish liability accounts for the unearned premium reserves (unex-
pired risk reserve) and the loss (outstanding claims) reserve (see Outreville
(1998 [114])). The unexpired risk reserve can be used to fund a surrender
value to the policyholder in the case of withdrawal from the insurance policy.
The loss reserve is the reserve to pay anticipated losses, including incurred
but not reported (IBNR) losses. In practice these losses create the com-
panys main liabilities. An insurance companys assets consist of its invested
funds arising from premiums and current value of surplus. These assets are
invested to generate income. The underwriting profit can be estimated by
deducting the current value of loss reserves and earned expenses from earned
Chapter 3 110

premiums. The insurance companys expenses may consists of acquisition

costs that are approximately proportional to premiums and administrative
costs that are approximately proportional to current loss reserves. On the
one hand the insurance liabilities (reserves) are increased by instantaneous
policy writing and at the same time are decreased by the instantaneous loss
payment outflow. On the other hand, assets are increased by instantaneous
earned premiums and the investment income inflow but are decreased by
instantaneous loss payments.
We will use the following notation:

St denotes the surplus of the insurance company at time t;

Lt denotes the expected current loss reserves at time t;

dPt denotes the instantaneous earned premium inflow at time t.

Let L be the underwriting profit and expense loading charged by the

insurer, expressed as a proportion of expected losses, and

dPt = (1 + L )Lt dt,

that is 1 + L is the gross insurance premium rate. Then the instantaneous

underwriting profit at time t follows the diffusion process

dt = dPt Lt dt L Lt dt P dPt + sL dWtL

= (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

dIt = rI (Lt + St )dt + I (Lt + St )dWtI .

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

+rI (1 + )St dt + L Lt dWtL + I (1 + )St dWtI

= 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

Similarly to the SR model we define the objective function to be max VTRBC .

Again, the insurer can adjust the value of risk to the optimal value
= arg max V0RBC by:

using reinsurance to control the underwriting risk L , and/or

using optimal investment strategies to control the investment risk I .

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.

3.4.2 Risk management incentives in the SR and RBC

models of the maximization of shareholders value
in the presence of FD costs

In this subsection we calculate the value of the objective function (insurer

value) in the SR and RBC models of the maximization of shareholders value
in the presence of FD costs. To illustrate the existence of risk management
incentives in insurance in the presence of FD costs, we provide an analytical
form of insurer value in the RBC model with FD costs that come in the
form of lost upside potential of the surplus. We show that the insurer value
is a function of insurance companys risk, FD barrier, initial level of capital-
ization, required minimal level of capitalization and parameter of FD costs.
We determine the optimal value of insurance companys risk that maximizes
the insurer value, and investigate the companys risk management sensitiv-
ities with respect to parameter of FD costs, FD barrier and initial level of
capitalization.
Chapter 3 113

The main results of the SR model

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 .

Now, using Levys Theorem on martingale characterization of Brownian mo-

tion (see Karatzas and Shreve (1988 [93]) and also see Lemma 1 in Appendix
A in Sherris and van der Hoek (2004 [136])) it can be shown that

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

dWtL dWt = L dt.

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)+

is the value at time 0 of a European call option on an underlying asset paying

a continuous dividend at rate r
e with current price 0 , exercise time T ,
and a strike price K. Using the classical Black-Scholes result we obtain

c0 = 0 e(re )T (d+ ) K erT (d ) , (3.4.9)

 
0 2
ln K
+ e 2 T

where d =
T
and is the standard normal cdf.
Chapter 3 115

Using (3.4.9) we can immediately calculate the expectation in the first

term in (3.4.8) and also the expectation in the second term, which is equal
to the first term in (3.4.9) with K = b.
So,
( " + # )
1
V0SR = L0 eL T w EQe erT T + (1 w)EQe erT T 1{T > b}
w

= L0 eL T w 0 e(re )T (h+ ) erT (h )

+(1 w)0 e(re )T (l+ )

= L0 w 0 ee T (h+ ) (h ) + (1 w)0 ee T (l+ ) , (3.4.10)
   
2 0 2
ln(w 0 )+ e 2 T ln b
+ e + 2 T

where h =
T
and l+ =
T

(we also used the fact that L = r).

From (3.4.10) we can see that the value V0SR depends on FD costs, the
FD barrier and the parameters L , L and .
1
Taking the partial derivative of V0SR with respect to w for b
< w < 1 we
obtain

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

investigate the FOC

SR
V = L L L0 T 0 ee T [w (h+ ) + (1 w) (l+ )]
0

e T h+
+L0 w 0 e (h+ ) (h ) h+ T

l+
+L0 (1 w)0 ee T (l+ )


e T
= L L L0 T 0 e [w (h+ ) + (1 w) (l+ )] + L0 T (h )
!
T ln b0 + T
+L0 (1 w) 0 ee T (l+ ) .
2 2
T
Taking into account the fact that a higher value of liabilities corresponds
to a lower value of the solvency ratio, we conclude that the instantaneous
correlation L between the liabilities and solvency ratio is negative. Under
this condition, the FOC equation
SR
V =0
0
may have a solution for optimal . The FOC equation is non-linear with
respect to the and can not be solved analytically. In order to obtain
results, numerical methods must be used to find an optimal value .

The main results of the RBC model

We consider the following three forms of financial distress costs:

1) deadweight losses are proportional to the terminal value of the insurers
surplus with proportionate coefficient 1 w, w (0, 1), and
2) deadweight losses are a constant amount C
3) deadweight losses are of the form of lost upside potential of terminal value
of companys surplus.
For the first form of FD costs the value of the function F at ST as the
terminal value of the companys excess of regulatory surplus net of FD costs
is equal to

F1 (ST ) = w ST ,
Chapter 3 117

for the second form it is

F2 (ST ) = ST C,

and for the third form it is

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

in the second case where F = F2

(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} ,

and in the third case where F = F3

h
(3)
V0RBC = V0 = e rT
EQ (ST S )+ 1{mST >D} (3.4.13)
i
+(ST S )+ 1{mST D}{ST S +U } + U 1{mST D}{ST >S +U } .

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

(see Etheridge (2002 [54]) pp. 144-148).

The value of V0RBC in (3.4.11) is

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

where () is the standard normal distribution function.

Chapter 3 120

Similarly, the second integral is

Z Z
1 y2
erT K e2ab pT (2a + bT, x) dx = KerT e2ab e 2 dy
2
x0 x0 2abT

T
2a+bT x0

Z
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

From here we obtain

h i D 2r2 1 D2
+
EQ (ST K) 1{mST D} = C , 0, K, T
S0 S0
2
2r2 1 2
D
ln KS0 + r 2 T 2
D D
=
S0 S0 T
2
D 2
ln KS0 + r 2 T
K erT ,
T

where C (x, t, K, T ) is the price at time t of a European call option with

strike price K, maturity time T and the stock price x at time t.
Therefore,
2r2 1 2 2
(1) D D D S
V0 = C (S0 , 0, S , T ) C , 0, S , T w C , 0, , T ,
S0 S0 S0 w

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

However, using change of the numeraire we can derive an analytical form

of optimal for some cases. To illustrate and support our initial hypothesis
of existence of insurance risk management incentives under the present of FD
costs we will evaluate the shareholders value for the third form of FD costs
and r = 0. According to the Numeraire Invariance Theorem (see Geman et al
(1995 [62]) or Duffie (2001 [50])) there exists a sequence of pairs: a numeraire
(i)
Nt = eri t (with ri+1 < ri for all i) and a probability
measure QN(i) , equiv-
f (ST ) f (ST )
alent to the initial measure Q, such that EQN (i) (i) = EQ (i) rT .
NT N0 e

Taking ri 0 we obtain a risk-neutral measure QN () = Q0 such that
h i
EQ0 [f (ST )] = EQ f e(SrTT ) .
h
(3)
V0 = EQ0 (ST S )+ 1{mST >D} + (ST S )+ 1{mST D}{ST S +U }
i h i
+U 1{mST D}{ST >S +U } = EQ0 (ST S )+ (ST S )+ 1{mST D}
h i
+ +
+EQ0 (ST S ) 1{mST D} (ST S ) 1{mST D}{ST >S +U }
h i
+EQ0 U 1{mST D}{ST >S +U } = EQ0 (ST S )+
h i
EQ0 (ST S ) 1{mST D}{ST >S +U }

+U Q0 {mST D} {ST > S + U }
h i
= EQ0 (ST S )+ EQ0 (ST (S + U )) 1{mST D}{ST >S +U }

S0 (S + U )
= C (S0 , 0, S , T ) C D, 0, ,T .
D

To find an extremum
b of the function V0 () first we have to calculate
Chapter 3 122

the derivative of C (x, t, K, T ) = x (d+ ) K (d ).

( 2
! 2
!)
ln Kx + 2 T ln Kx 2 T
C (x, 0, K, T ) = x K
T T
d+ d
= x (d+ ) K (d )


d+ d+
= x (d+ ) K (d ) T


d+ d+
= x (d+ ) x (d+ ) T

2
!
ln Kx + 2 T
= x (d+ ) T = x T (3.4.14)
T
(d ) x
where we have used the relationship (d+ )
= K
.
Using (3.4.14) we obtain
! 2 2
!
(3)
2
ln SS0 + 2 T ln S0 (SD +U ) + 2
T
V = S0 T D T.
0 T T

(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

b2 w.r.t. the FD D gives

audits. Taking the first-order derivative of

2 2 D2 D2 S

b = ln + ln 2
D D T ln S S+U
S (S + U ) S0 (S + U )

we conclude that D
b2
< 0 since according to the model setup D < S S0 .
This means that when the FD barrier D increases, the company becomes
more sensitive to the FD costs. Since such costs are caused by assumed risk,
the companys risk managers will then tend to reduce such risk.
It is obvious that the optimal companys risk is inversely proportional to
the length T of the observed period.
b2 w.r.t. the companys initial level
Taking the first-order derivative of
S0 of capitalization
2 2 D2

b = ln
S0 S0 T ln S S+U
S (S + U )

we see that S0
b2
> 0 since D < S . This means that the higher the
companys initial level of capitalization the more risk it is willing to assume.
And finally, more importantly, we consider the companys risk manage-
ment sensitivity with respect to the parameter U of FD costs.
S02 (S +U ) !
2 1 ln D2 S
ln S (SD2+U )

b =
U U T ln S S+U


S 2 (S +U ) S 2 (S +U )

1 ln 0D2 S + ln S (SD2+U ) ln S S+U

ln 0D2 S ln S (SD2+U )
=
T (S + U ) ln2 S S+U


S0 S (S +U )
2 ln D2 + ln S ln S S+U
2 ln D0 + ln (S S+U
S

)
2 ln SD + ln (S S+U

)

= 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 .

i.e. it depends on variable parameter of proportional reinsurance and

parameter of investment risk I . Therefore, the companys risk managers
can control the integrated risk (, I ) in a two-dimensional way:

control ex-post underwriting risk through reinsurance (by choosing );

control investment risk I through risk management of its investment

portfolio.

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

Enhancing insurer value via

reinsurance and dividend
dynamic optimization: multiple
period models

In this chapter we investigate the stochastic control problems of finding

optimal reinsurance (insurance risk transfer) and dividend-payment strate-
gies in the maximization of shareholders value in a dynamic setting. The
demand for reinsurance is investigated in models of the risk process with
dividend barrier strategies in a frictionless environment and in the presence
of insolvency costs.

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

actuarial science F. Lundberg is considered to be a pioneer in introducing

and developing a dynamic theory of the insurance process (collective risk
theory) which models the insurers risk reserve (premium income minus in-
surance claims) as a time dependent stochastic process. Some elements of
this theory were first presented in his doctoral dissertation at the University
of Uppsala (Sweden) in 1903, and then later a complete development of the
theory was presented at the International Congress of Actuaries in Vienna
(1909 [99]). Lundbergs theory was essentially developed later by the famous
Sweden statistician and actuary H. Cramer (1930 [32] and 1955 [33]) and
is based on calculation of non-ruin probability (or survival probability) of
the insurance company, that is, the probability that a dynamic risk reserve
process of the company remains positive (the company remains solvent) dur-
ing the specified (finite or infinite) period of time. The survival (non-ruin)
probability gives a measure of reliability of the contracts the insurance com-
pany has underwritten, but it does not reflect the level of profitability of the
insurance business. In the real world insurers, as agents that act on behalf of
their principals (shareholders), strive to provide their shareholders with a fair
return on risk capital supplied, and at the same time as insurance debt oblig-
ors of their debtholders (policyholders) must carry underwriting risk against
compensation for policyholders. Therefore, the main insurer objective is to
create shareholders value and remaining solvent (obligation against policy-
holders) is an important but secondary objective, the realization of which is
usually monitored by a regulator. De Finetti, the prominent Italian actuary,
was the first one who seriously revised Lundbergs collective risk theory and
transformed it to the new dynamic and economically interesting approach.
In his paper presented at the XV International Congress of Actuaries in New
York in 1957 (1957 [44]) de Finetti noticed that the ultimate survival proba-
bility of the company would be zero, unless the company allowed the reserve
Chapter 4 128

capital to grow without limit. Obviously, it is not realistic that an insur-

ance company can operate in a way which allows unlimited accumulation of
capital, since it has to pay dividends to its shareholders when the insurers
reserve capital increases to provide them with a fair return on the risk cap-
ital they supplied. De Finetti suggested that the companys objective, as
an alternative to maximization of the non-ruin probability (or minimization
of the ruin probability), may be to find the dividend-payment policy (divi-
dend cash flow) which maximizes the expected discounted sum of all future
dividend payments. This suggestion has actually opened the door for the
application of modern stochastic control theory to problems in insurance.
Later the problem of finding the optimal dividend-payment policy has been
investigated extensively by Karl Borch (1974 [17]; 1990 [19]), H. Buhlmann
(1970 [25]), and H. Gerber 1972 [63]; 1979 [64]. It should be noticed that the
criterion of maximizing the discounted value of expected future dividends
is consistent with current theories in modern financial economics, where ac-
cording to theorem of Miller and Modigliani (1961 [104]) the value of a firm
is equal to the present value of the net distributions to shareholders over
an infinite horizon. Although the original Miller-Modigliani theorem was
formulated under the assumption of perfect certainty, there exist several
extensions of this work confirming the validity of the Miller-Modigliani ap-
proach in the case of a stochastic environment (e.g. Sethi et al (1984 [132];
1991 [133]), and Sethi (1996 [131])).
The standard framework for the problem of finding the maximum of
the discounted value of expected future dividends over the period from the
current point of time to the ruin time is to model the insurer risk reserve as
a controlled piecewise deterministic Markov process (PDMP) or a controlled
diffusion with a control variable defined by dividend-payment policy. Little
research has been done on this stochastic control problem under a PDMP
Chapter 4 129

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

It is worth noticing that maximization of the expected value of discounted

dividend-payments until the time of ruin is indeed important, but it is not
a complete analysis of shareholders value creation. This is because the ex-
pected 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 as the com-
panys investors would prefer to measure the investment performance of the
supplied equity capital through the risk-adjusted discount rate, under which
the shareholders NPV of their investment project is zero. The risk-adjusted
discount rate is an effective measure of shareholders value creation. In this
chapter we explain the relationship between the two notions: expected pre-
sented value of future dividends and risk-adjusted discount rate.
This chapter is organized as follows. In Section 4.1 we overview early
contributions to the research area of finding the optimal reinsurance and
dividend-payment strategies in a dynamic multi-period setting. In Section
4.2, we analyze stochastic optimal control problems of finding an optimal
quota share proportional reinsurance and dividend-payout policies for risk
reserve (surplus) process of the insurance company described by a piecewise
deterministic Markov process (PDMP) and/or diffusion process. In Shreve
et al (1984 [137]) the authors showed that for general diffusion models of
surplus the optimal dividend-payout policy yields a barrier. This general re-
sult confirms the correctness of all early intentions of De Finetti (1957 [44]),
Buhlmann (1970 [25]) and Gerber (1979 [64]) to optimize dividend-payout
strategies with a barrier. In Section 4.3 we investigate the expected present
value of future dividends (shareholders value) in general Ito diffusion mod-
els of companys surplus with dividend barrier strategies and analyze the
demand for reinsurance in these models. We reconsider the same models
Chapter 4 132

under the solvency restrictions in Section 4.4, and in the presence of insol-
vency costs in Section 4.5.

4.2 Optimal reinsurance and dividend-payout

strategies in the maximization of share-
holders value: a stochastic optimal con-
trol approach
In this section we survey different models of dynamic insurance risk man-
agement, 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. The optimal control feedback provide us
with the maximal expected value of future discounted dividend payments.
This value as a function (value function) depends on the state variable, i.e.
initial reserve, and the exogenous risk-free discount rate. We investigate the
properties of the value function and analyze the relation between it and the
risk-adjusted discount rate.

4.2.1 Mathematical modelling of insurer surplus

Surplus as a Levy process: general setup

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

We admit the possibilities of taking a change-loss reinsurance by the direct

insurer, and as in the previous chapter I1a, b (Y ) = Y (1 a) (Y b)+ de-
note the retained1 part of the underwritten claim. It is assumed that both
the cedent and the reinsurer use a mean value premium principle. Assum-
ing the insurance and reinsurance premiums are meant to be arbitrage-free
we get the following necessary conditions of arbitrage-free pricing in insur-
ance: the reinsurer is more risk averse, and thus the reinsurance risk (safety)
loading is greater than or equal to the analogous risk loading corresponding
to the direct insurance price. Moreover, it has already been shown in the
preceding chapter that for change-loss reinsurance contracts the reinsurance
loading is proportional to insurance risk loading with proportionality
R

(1F (y)) dy
b
1+
factor k(b, ) = R
1.
(1F (y)) dy
b
Now, it follows from the assumptions established above that the premium
rate is

c = (1 + ) m (1 + k(b, )) [m E [I1a, b (Y )]]

= E [I1a, b (Y )] + (m k(b, ) [m EI1a, b (Y )])

1
In this chapter a stands for the cedents retention level of the proportional component
of change-loss reinsurance.
Chapter 4 134

In the case of quota share proportional reinsurance (i.e. b = 0) k(0, ) = 1

and therefore c = a (1 + ) m.
Suppose that in addition to this there is a debt that the company pays
to its fixed claimants at the constant rate . Those payments cannot be
controlled.
Further, it is assumed that the company pays dividends to its sharehold-
ers. Let Lt denote the cumulative amount of dividend paid-out up to time t.
It is required that Lt is Ft -adapted and is a non-decreasing right-continuous
with left limits process (cadlag). It is allowed the following representation
of Lt
Zt
Lt = lt dt,
0

where lt denote the density of dividend distribution (dividend rate).

Finally, it is assumed that the company has an option to choose the
amounts and times of the dividend distribution and reinsurance. This means
that the company can dynamically control its business by two control vari-
ables: reinsurance and dividend payments. We describe control by a three-
dimensional stochastic process {a (t), b (t), l (t)}, where 0 a (t) 1
corresponds to the retained fraction (cedents retention level) of quota share
proportional reinsurance contract, b (t) 0 corresponds to the retention
level of excess-of-loss reinsurance, and l (t) 0 is the density of dividend
distribution.
Summarizing all the above assumptions, we can write down the dynamics
of the risk reserve (surplus or net-worth) process of the company under the
control policy

dRt = (c(a (t), b (t)) + rI Rt l (t)) dt (4.2.1)

+I Rt dW I dSt (a (t), b (t))

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

V (x) = V (x) = sup V (x). (4.2.4)

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.

The piecewise deterministic Markov model of the companys sur-

plus

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.

Diffusion approximation of the companys surplus

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

Note that in the case of quota share proportional reinsurance =

a m = a and 2 = a2 (s2 + m2 ) = a2 2 , and in the case of excess-of-loss
Rb Rb
reinsurance = F (y) dy and 2 = yF (y) dy.
0 0
Therefore, assuming the independence of WtS and WtI the risk reserve
process can be approximated by the following diffusion

dRt = ((a (t), b (t)) + rI Rt l (t)) dt (4.2.6)

q
+ 2 (a (t), b (t)) + (Rt )2 I2 dWt

dR0 = x, (4.2.7)

where Wt is a Ft -adapted standard Brownian motion.

It is difficult to find the optimal control for the general diffusion form
(4.2.6) of risk reserve. However, it is possible to find an explicit form of
optimal control in the following particular cases:

Control by quota share proportional reinsurance and dividend pay-

ments in the absence of investments and debt liability (i.e. = rI =
I = 0 = b);

Control by excess-of-loss reinsurance and dividend payments in the

absence of investments and debt liability (i.e. = rI = I = 0 = a);

Control by quota share proportional reinsurance and dividend pay-

ments in the presence of investments and absence of debt liability (i.e.
= 0 = b);

Control by quota share proportional reinsurance and dividend pay-

ments in the presence of debt liability and absence of investments (i.e.
rI = I = 0 = b).
Chapter 4 138

4.2.2 Solutions to the particular stochastic control mod-

els of insurers surplus

Controlling the insurers underwriting risk and dividends payout:

example of quota share proportional reinsurance

In this particular case the dynamics of the companys risk process are given
by

dRt = (a (t) l (t)) dt + a (t) dWt (4.2.8)

dR0 = x (4.2.9)

The corresponding value function is given by

Z
V (x) = E et lt dt,
0

where = inf{t : Rt 0}, and the optimal value function is defined as

V (x) = sup V (x),

where is the set of all admissible controls.

To find the optimal control we will use classical stochastic control
theory (for the theory see Fleming and Rishel (1975 [56]), Fleming and Soner
(1975 [57]), or Krylov (1980 [95])). Before dealing with the analytical part
of the problem, we show that the optimal value function V is nonnegative
and concave, and derive the corresponding Hamilton-Jacobi-Bellman (HJB)
equation, which the function V must satisfy.

Proposition 4.2.1. The function V is a nonnegative concave function.

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

az (t) = ax (t) + (1 ) ay (t)

Chapter 4 139

and

lz (t) = lx (t) + (1 ) ly (t) (4.2.10)

Linearity of equation (4.2.8) implies that z is an admissible control for the

initial reserve z = x + (1 ) y and Rtz = Rtx + (1 ) Rt y with ruin
time z = min{x , y }. Also, linearity of the value function implies

Vz (z) = Vx (x) + (1 ) Vy (y).

Now, > 0 we choose the controls x and y such that V () V () ,

and obtain
V (z) Vz V (x) + (1 )V (y) .

From here we obtain the concavity of V when tends to 0.

We will derive the HJB equation for two different cases:

1) the rate of dividends payout l is bounded by a constant B;

2) the rate of dividends payout is unbounded.

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

For any admissible control and u [0, x] consider Tu = min{u, inf{t :

a.s.
Rt
/ (x u, x + u)}}. It is obvious that Tu < and Tu 0 as u 0.
Consider a particular control
b such that for all t 0 ab (t) = a and lb (t) = l,
where a and l are arbitrarily fixed constants. For such particular suboptimal
control
b we obtain
u
ZTb
u
V (x) E es l ds + eTb V RTbbu .
0

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]

On the other hand 1 > 0 there exists a control

e such that
Tu
Z
u
V (x) = sup E es l (s) ds + eT V RTu

0
u

T
Ze

u
E es le (s) ds + eTe V RTeeu + 1 .
0
Chapter 4 141

By choosing 1 = E2 [Teu ] and applying Itos Lemma we obtain

u
T
Re
E es le (s) + A(ae (s), le (s)) V Rse ds + E2 [Teu ]
0
0
E [Teu ]
u
T
Re
E es max l + A(a, l) V Rse ds + E2 [Teu ]
0 a[0,1]; l[0,B]

E [Teu ]
u0
max l + A(a, l) [V (x)] . (4.2.14)
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)

The general solution to the latter differential equation is V (x) = C1 x , where

= 2
.
+
2 2
x 2
From here we get a (x) = 2 (1)
, and conclude that x0 =
(1 ).
Solving HJB equation (4.2.11) for all x [x0 , x1 ) (i.e. a (x) = 1) we obtain
the following solution

V (x) = C2 e ()x + C3 e+ ()x ,

2 +2 2
where () = 2
.
If x > x1 , then l (x) = B, a (x) = 1 and any concave solution of HJB
equation in 2) is given by

B
V (x) = C4 e (B)x + .

Therefore, the solution to (4.2.11) is given by

C1 x , x < x0

V (x) = C2 e ()x + C3 e+ ()x , x [x0 , x1 )


C4 e (B)x + B , x x1

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

V (x0 ) = V (x0 +); V (x1 ) = V (x1 +)

V 0 (x0 ) = V 0 (x0 +); V 0 (x1 ) = V 0 (x1 +) = 1.

Chapter 4 143

Solving these five equations we obtain

1 ( B) ()
x1 = x0 + ln ,
+ () () + () ( B)

C1 = 2 + () () 2 e ()(x1 x0 ) + e+ ()(x1 x0 ) ,
e ()x0
C2 = ,
() (e ()(x1 x0 ) + e+ ()(x1 x0 ) )
e+ ()x0
C3 = ,
+ () (e ()(x1 x0 ) + e+ ()(x1 x0 ) )
e (B)x1
C4 = .
( B)

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.

In summary, we conclude that the optimal value function V is equal to

b
C1 x , x < x0

V (x) = b2 e ()x + C
C b3 e+ ()x , x [x0 , x1 )


x+C b, xx
4 1

Using boundary conditions

V (x0 ) = V (x0 +); V (x1 ) = V (x1 +)

V 0 (x0 ) = V 0 (x0 +); V 0 (x1 ) = V 0 (x1 +) = 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.

Controlling the insurers underwriting risk and dividends payout:

an example of excess-of-loss reinsurance

This particular stochastic problem was investigated by Asmussen, Hjgaard

and Taksar (2000 [9]). In this case the dynamics of the companys surplus,
Chapter 4 146

given by (4.2.6), are reduced to

dRt = ( (b (t)) l (t)) dt + (b (t)) dWt (4.2.17)

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 :

1) in the case of bounded dividend rate

(v) 00 0
0= max V (x) + (v l) V (x) V (x) + l
v[0, ],l[0,B] 2

2) in the case of unbounded dividend rate

(v) 00 0 0
0 = max max V (x) + v V (x) V (x) , 1 V (x)
v[0, ] 2

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

1.1) Bounded dividend rate and infinite support.

The optimal value function is
x
R1 1

R
x dz
ey
b(v(z))
dy, x < x1
V (x) = 0

e
B + e(v1 )(xx1 ) , x x1 ;
e )
(v 1

(vB) (vB)2 +2 (v)
e
where (v) = (note b(v) is the inverse function of
(v)

(b)); the value v1 is the root of the following functional equation

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

1.4) Unbounded dividend rate and finite support.

The optimal value function is
x
R0 1

Rx dz


b(v(z))
C e dy, x < x0
y

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 .

Controlling the insurers underwriting risk and dividends payout

in the presence of investments

In this subsection we consider the problem of optimal dividends payout and

optimal risk control via quota share proportional reinsurance under the as-
sumption that the companys surplus is instantaneously invested in a finan-
cial assets with the price governed by a geometric Brownian motion. In this
particular case of stochastic control the dynamics (4.2.6) of the companys
Chapter 4 150

surplus can be rewritten in the following way

dRt = (a (t) + rI Rt l (t)) dt (4.2.21)

q
+ a (t) 2 + (Rt )2 I2 dWt

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

Controlling the insurers underwriting risk and dividends payout

in the presence of debt liabilities

In this particular case of stochastic control the dynamics (4.2.6) of the com-
panys surplus can be rewritten in the following way

dRt = (a (t) l (t)) dt + a (t) dWt (4.2.23)

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

which is an increasing function.

4.2.3 The analysis of the value function under the op-

timal control policy: risk-adjusted discount rate
and the companys level of capitalization

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 )

R(x) > V 0 (x) > V 0 (x1 ) = 1.

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

future dividends under the risk-adjusted probability measure) is equal to the

amount of companys risk capital provided by shareholders. So, the value
function is constant (it is equal to x) under the risk-adjusted probability
measure, and thus the maximization of the value function is reluctant.
On the other hand the optimal value function under the physical mea-
sure is useful when we construct the corresponding risk-adjusted probability
measure. Indeed, let us consider the new probability measure Q such that
dQ 1 ln R(x)
dP
= R(x)
. Then for (x, t) = t
, x (0, x1 )

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:

is a decreasing function of x (i.e. a lower level of company capital-

ization corresponds to a higher risk loading loading on the risk-free
rate of return).;

is a decreasing function of the risk-free rate (i.e. a higher value of

the risk-free rate corresponds to a lower value of its risk loading ).
Chapter 4 155

4.3 Reinsurance and dividend optimization

in uncontrolled dividend-barrier models
of surplus
It is evident in solving most of the stochastic control problems from the
previous section that the optimal dividend policy takes the form of a bar-
rier policy. This pattern is perhaps due to the general result obtained by
Shreve et al (1984 [137]). The authors considered the problem of optimal
consumption for general diffusions, and showed that under some reasonable
assumptions the optimal policy yields a barrier b , such that whenever the
companys surplus exceeds b the excess is immediately paid out as divi-
dends.
It should be mentioned that in fact, the optimality of a barrier dividend
policy was revealed much earlier during the investigation of the discrete-
time counterpart of this problem. De Finetti (1957 [44]) was the first to
consider a discrete time risk model in the situation where the aggregate gain
of an insurer per unit time is either 1 or 1 and showed that the optimal
dividend policy is to pay an excess of companys surplus over a specific
level to shareholders as dividends. Morill (1966 [106]) who also studied
the conditions under which such a policy is optimal, called this policy a
barrier policy. The problem of optimal dividend strategies has also been
considered in continuous time within the Cramer-Lundberg framework (see
textbooks by Buhlmann (1970 [25]) and Gerber (1979 [64])), and within
diffusion framework (see Gerber and Shiu 2004 [67]).
In this section we consider uncontrolled stochastic dynamic programming
problem of maximization of future discounted dividend payments within
the general Ito diffusion framework. This means that the form of dividend
policy is predetermined, and it is meant to be a barrier policy with a given
Chapter 4 156

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])).

4.3.1 General results for Ito diffusion model

It is assumed that the uncontrolled surplus follows the Ito diffusion

dRt = (Rt )dt + (Rt ) dWt

R0 = x,

where W is a Brownian motion and and are Lipschitz continuous. When

the companys surplus reaches its lower bound, i.e. zero, then the company
is ruined and it exists from the market. This lower bound at zero is usually
considered as an absorption barrier.
Chapter 4 157

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

Lt = (MR (t) b)+ ,

where MR (t) = max Rs .

0st
Taking into account the dividend payments we obtain the reflected (by
dividend barrier) diffusion process of the companys surplus

dRt = (Rt )dt + (Rt ) dWt dLt

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

In addition it is assumed that the value function V is twice continuously

differentiable. With this assumption we can construct the corresponding
differential equation which the value function V must satisfy. We can do
this using the infinitesimal generator (differential operator) of Ito diffusion
for all x (0, b)
E [V (Rt , b) V (x, b)]
AV (x, b) = lim
t0 t
E [(1 + t + o(t))V (x, b) V (x, b)]
= lim = V (x, b).
t0 t
According to the representation theorem for the generator of Ito diffusion
(see ksendal (2000 [112]), p.117) we obtain
1
AV (x, b) = (x)V 0 (x, b) + 2 (x) V 00 (x, b).
2
Chapter 4 158

Therefore the value function V satisfies the following differential equation

1 2
(x) V 00 (x, b) + (x)V 0 (x, b) V (x, b) = 0, 0 x b (4.3.1)
2
with two boundary conditions. If the initial surplus is zero the company is
bankrupt, and no dividends are paid, i.e. V (0, b) = 0. The second boundary


condition is x V (x, b)x=b = 1. This follows from the relationship
V (b + , b) V (b, b)
lim = lim = 1.
0+ 0+

To solve (4.3.1) in addition to the Lipschitz conditions on drift and diffusion

coefficients we impose the following restriction on the drift

0 (x) , x 0.

This assumption allows us to solve (4.3.1) much easier using traditional

methods of ODE. In general, when this assumption is violated, it can be
shown that the solution of (4.3.1) has very complex behavior, which may
lead to the violation of smooth-pasting conditions.
Now, using the arguments of the Lemma 4.1 in Shreve et al. (1984 [137])
we can easily derive the following result

Proposition 4.3.1. Let g be a solution of the equation

1 2
(x) g 00 (x) + (x)g 0 (x) g(x) = 0, x 0 (4.3.2)
2
with g(0) = 0 and g 0 (0) > 0. Then
1) g 0 (x) > 0 for all x;
2) if there exists b so that g 00 (b ) = 0, then (x b )g 00 (x) 0 for all x.

Note that boundary conditions for g at 0 uniquely determine the solution

g. If g is a solution of (4.3.2) in Proposition 4.3.1 with g(0) = 0 and g 0 (0) > 0,
then using the second boundary condition for the value function V 0 (b, b) = 1
we get the following form of V for 0 x b
g(x)
V (x, b) =
g 0 (b)
Chapter 4 159

From here we conclude that the value function V is maximized at b <

(if such exists) using a dividend policy with a barrier, where

g 00 (b ) = 0.

If no such b < exists, then using part 2) of Proposition 4.3.1 we conclude

that g 00 (x) is concave for all x 0, that is, g 0 is a decreasing function and
the value function is maximized at b = , which means that no optimal
dividend barrier policy exists.
If the optimal dividend barrier b is finite, then it follows from part 2)
of Proposition 4.3.1 that g 00 (x) is a concave function for x b , and thus
V (x, b) is also a concave function for 0 x b b .

Examples

Example 1. In this example we consider the diffusion approximation of the

companys surplus by an arithmetic Brownian motion, which can be obtained
as a result of diffusion approximation of the classical Cramer-Lundberg risk
model. Such an Ito diffusion model of the companys surplus with positive
constant drift and diffusion coefficients has been studied by Shreve et al.
(1984 [137]), Jeanblanc-Picque and Shiryev (1995 [89]), Assmusen and Tak-
sar (1997 [8]), and more recently by Gerber and Shiu (2004 [67]). In this case
0 (x) = 0 < and the reflected diffusion process of the companys surplus
is

Rt = R0 + t + Wt Lt .

The corresponding ODE which the value function V must satisfy is

1 2 00
V (x, b) + V 0 (x, b) V (x, b) = 0, 0 x b.
2
From here we find the function g

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

Solving g 00 (b) = 0 we find the optimal barrier

ln | | ln +
b = 2 <
+

Example 2. In this example we consider the diffusion approximation of

the companys surplus by a general geometric Brownian motion. We let the
drift to be (x) = x + and the diffusion to be (x) = x + = ( x + )
with > = 0 (x) > 0 and > 0. The function g is of the form

g(x) = C1 ( x + )1 + C2 ( x + )2 , x [0, b]

where 1 > 0 > 2 are the roots of

1 2 2 2 1 2 2
z + z = 0.
2 2

Therefore, the optimal value function is of the form

(

C1 ( x + )1 + C2 ( x + )2 ; 0 x b ;
V (x, b ) =
x b + V (b , b ), x > b .

Using both boundary conditions and g 00 (b ) = 0 we find

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

Example 3. In this example we consider the diffusion model of the com-

panys surplus in which the surplus is instantaneously reinvested in financial
instruments with a price process governed by a geometric Brownian motion
Chapter 4 161

(see equation (4.2.21) in the preceding section). The corresponding ODE

which the value function V must satisfy is

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.

4.3.2 Analysis of the value function and the optimal

dividend barrier in Ito diffusion model of surplus

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 }, ];

concave on [b, b ] if b < b ;

convex on [b , b] if b > b ;

Using concavity of V on [0, bb ] we can analyze the relationship between

the value function and the initial value of surplus (equity capital supplied
Chapter 4 162

by shareholders). Consider the value function per unit of initial surplus

V (x, b) V (x, b) V (0, b)
= ,
x x0
which can be interpreted as the slope of a secant. It follows from the concav-
V (x,b)
ity of V that the ratio x
is a decreasing function of x on [0, b] [0, b ].
V (b,b)
In particular, when x = b, this ratio is b
> V 0 (b, b) = 1. Therefore,
we get V (x, b) > x for all x [0, b] [0, b ]. Finally, if x > b then
V (x, b) = x b + V (b, b) > x.
V (x,b)
The same inequality for the ratio R(x, b) = x
> 1 was obtained by
Gerber and Shiu (2004 [67]), who considered the problem of optimal dividend
barrier for the particular case of the Ito diffusion of surplus process where
and are constant. This result is somewhat strange at first glance. Indeed,
the inequality V (x, b) > x means that there is no such discount rate > 0
under which the expected present value of dividend payments until the ruin
time is equal to the capital (initial surplus) supplied by shareholders. Also
Gerber and Shiu (2004 [67]) argue that the fact that R(x, b) is a decreasing
function of the initial capital x has a somewhat shocking implication: If
the investor is only interested in the leverage ratio (R), he or she would
want to invest in companies with a low degree of capitalization!. We take
an alternative view on this, and try to explain it from a financial economics
point of view. First of all, one can notice that the value function is defined by
the expectation with respect to the physical measure, and thus the discount
rate is not adjusted for risk (it is lower, or without risk loading). If we
dQ 1
consider a new risk-adjusted measure Q such that dP
= R(x,b)
, then
(x) (x)
Z Z
1
e(+ t )t lt dt .
ln R(x,b)
x= V (x, b) = EQ et lt dt = EP
R(x, b)
0 0

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.

Proposition 4.3.2. Let and are Lipschitz continuous and (0 x) .

Then for b > (<)b where b <

(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

4.3.3 Reinsurance optimization in an uncontrolled div-

idend barrier model of surplus: an example of
the model governed by an arithmetic Brownian
motion

Consider a diffusion model of the companys surplus defined by an arithmetic

Brownian motion with constant drift and diffusion coefficients. It is assumed
that at time 0 the insurer may decide to cede part of its underwriting risk
using a quota share proportional reinsurance with retention level a (0, 1).
Then the new drift and diffusion coefficients are a and a respectively, and
the optimal dividend barrier is

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

in the risk-adjusted valuation

(x,a) (x,a)
Z Z 
ln R(kx,b )

+ t
x = EQ et lt dt = EP e t
lt dt
0 0

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.

4.4 Reinsurance and dividend optimization

in uncontrolled dividend-barrier models
of surplus with solvency restrictions
In this section we reconsider the problem of reinsurance and dividend op-
timization in uncontrolled dividend-barrier models of the insurers surplus
governed by Ito diffusions in the presence of solvency constraints. Since
the policyholders (insureds) accumulate their credit exposure with one or
few insurance companies, they are thus sensitive to the financial strength
of the insurer, where the financial strength is determined by solvency con-
straints imposed by a regulator. To satisfy these regulatory requirements
insurers recognize the need for security by holding risk capital, which is
usually supplied by shareholders. Now the higher value of risk capital, the
more underwriting risk the insurer wants to assume. But at the same time
a highly leveraged insurer is required to hold more risk capital to satisfy
solvency restrictions.
The solvency restrictions are defined by a holding risk capital at the level,
Chapter 4 167

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

4.4.1 Optimal dividend barrier policy and reinsurance

under solvency restrictions

We consider a reflected (by a dividend barrier) Ito diffusion process of the

companys surplus

dRt = (Rt )dt + (Rt ) dWt dLt

R0 = x

with the aggregate dividend payments until time t

+
Lt = max Rs b .
0st

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

can be represented by the following formula

g(x)
V (x, b) = ,
g 0 (b)

where the function g satisfies the equation (4.3.2), i.e.

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.

Proposition 4.4.1. If the optimal dividend barrier b of unconstrained prob-

lem of dividend maximization is less than the regulatorys boundary b, then
the optimal dividend barrier under solvency constraints is equal to b.

This result was initially obtained by Paulsen (2003 [116]). A modified

proof of this result is provided below.

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)

where Lc is the continuous part of L. If Rs 6= 0 then Rs and Rs are greater

than b and thus V Rs , b V Rs , b = Rs = Ls and V 0 (x, b) = 1 for
all x [Rs , Rs ]. Therefore,

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

(x) (b) (x b) = (x b + V (b, b)) V (b, b),

Chapter 4 171

which is equivalent to

(x) (x b + V (b, b)) (b) V (b, b) 0.

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 .

Using standard methods of stochastic analysis it can be shown that the

survival probability as a twice continuously differentiable function of x and
Chapter 4 172

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

with boundary conditions

(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

use the method of separation of variables which is well-known in the theory

of PDE (e.g. see Polyanin and Zaitzev (2003 [120])).
It is worth noticing that the reinsurance has the special role of a vari-
able that controls the value of underwriting risk. Taking out reinsurance
decreases the value of the underwriting risk, and it thus decreases the value
of ruin probability, or equivalently decreases the level of regulators bound-
ary (minimal level of risk capital). At the same time the optimal dividend
barrier b will also decrease with purchasing reinsurance. So, taking out the
reinsurance may change the set of admissible dividend-barrier policies under
the solvency restrictions in such a way that b > b and thus b would be the
optimal barrier.

4.4.2 Optimal dividend rate, leverage ratio and rein-

surance in the presence of insolvency cost: an
expected utility approach

In the preceding subsection we introduced the solvency restrictions by means

of a certain level of ruin probability, and considered the problem of max-
imization of the shareholders value (the expected present value of future
dividends (consumptions)) for the uncontrolled dividend-barrier model of
the insurers surplus under the restriction on ruin probability (solvency re-
strictions). Essentially we considered a constrained problem of maximization
of future consumptions subject to the constraint on risk which is measured
by the probability of insolvency. It has already been argued at the beginning
of Chapter 3 that the main insurers objective is to maximize the sharehold-
ers value, and that without any constraints on underwriting risk the insurer
would behave as a risk-neutral agent. But as we know insurers do operate in
a regulatory environment with risk constraints. This is because the insurers
Chapter 4 174

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

rate (coefficient of proportionality to the surplus), leverage ratio and re-

tention level of quota share proportional reinsurance, for the surplus which
follows a geometric Brownian motion within the RBC solvency model of the
insurer considered in Chapter 3.

Insurer preference ordering under solvency constraints: a deriva-

tion of the corresponding utility function

Consider a discrete time ruin (Cramer-Lundberg) model of the surplus for

which the following equalities hold

Rt = Rt1 + c (St St1 ) = Rt1 + c St1 , t = 1, 2, ...

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]) ,

which is equivalent to p0 = E[S].

If we differentiate both sides of (4.4.2) with respect to t and then set
t = 0, we get

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

or p1 U 0 (x) = 0, which gives us p1 = 0.

Finally, by differentiating twice both sides of (4.4.2) and setting t = 0,
we get

0 = E [2p2 U 0 (x)] + E (p1 Y )2 U 00 (x)

= 2p2 U 0 (x) + Var [Y ] U 00 (x) ,

or equivalently p2 = 12 r(x) Var [Y ].

So, for small t we can write

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

and thus we get another approximation of P

1
P E[S] + r(x) Var [S] .
2

Comparing this with the first approximation of P we get

|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

Maximization of the shareholders value using utility approach

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

dRt = Rt dt + Rt dWt lt dt, (4.4.3)

where = (a, ) = a (L (1 P ) (L + P ) + rI pre (1 a)) + rI ,

p
= (a, ) = a2 2 L2 + (a + 1)2 I2 , W is a standard Brownian mo-
tion associated with the surplus, and lt is the dividend rate. The quantities
a, , L , P , rI , L , I are as defined earlier in Chapter 3. For convenience,
they are repeated here

a - insurers retention level of quota share proportional reinsurance

price rate of which is pre ;

1 + L - gross insurance premium rate;

- leverage ratio;

L , P - the expense ratios for administrative expenses (proportional

to losses) and for acquisition expenses (proportional to premiums) re-
spectively;

rI , I - drift and diffusion of log-return of stock market instrument

(the companys assets are instantaneously invested in stock market
instruments);

L - diffusion of loss reserve.

In addition it is assumed that the dividend rate is proportional to the surplus

with constant coefficient of proportionality, i.e. lt = Rt . Let the regulators
boundary for the insurers surplus be x < R0 = x, so that the insurer will
Chapter 4 179

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

g(a, , ) e(1m) y and substituting it to the ODE we get

1 2 2 (1m) y 1 2
g(a, , ) (1 m) e + g(a, , ) (1 m) e(1m) y
2 2
(1m) y (ey )1m
g(a, , ) e = ,
1m

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

V (x; a, , ) = C+ x + g(a, , ) x1m .

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

Now to find optimal uncontrolled variables such as retention level of quota

share proportional reinsurance, leverage ratio and dividend rate one maxi-
mizes the value function V to obtain

(a , , ) = arg max V (x; a, , ).

a(0,1); >0; (0,1)

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

In this chapter we consider and analyze different equilibrium pricing models

of insurance. A special attention is paid to the model of dynamic Wal-
rasian equilibrium in the integrated insurance-reinsurance market consisting
of the insureds (policyholders), direct insurers and reinsurers. The results
of maximization of shareholders value in a dynamic setting obtained in the
preceding chapter are used here to determine supply and demand sides of
insurance and reinsurance, and then find equilibrium insurance and rein-
surance prices under which the Pareto optimality is achieved. This chapter
includes numerical results which illustrate the existence of equilibrium in-
surance and reinsurance prices.

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

such as bonds. Therefore, in general, the optimal decision-making processes

for the insurance company are more realistic for dynamic-continuous analytic
frameworks than single-period frameworks.
In the past decade, several researchers, including Venezian (1994 [146]),
Chen and Venezian (1994 [27]) and Lin and Powers (1999 [98]) considered
dynamic insurance market equilibrium in a Walrasian framework. For in-
stance in Lin and Powers (1998 [98]), the authors constructed multi-period
models of insurance supply and demand in a market with homogeneous in-
sureds and homogeneous insurers. They did this by fixing the insurance
price and selecting the quantity of insurance for demand and supply sides
separately, and then provided a dynamic solution for the equilibrium price
and quantity. Here the equilibrium price was found so that aggregate de-
mand is equal to aggregate supply in insurance market consisting of insureds
and insurers. And according to Borch (1962 [16]) the risk exchange between
insureds and insurers under such equilibrium price is Pareto optimal.
In this chapter we extend the Lin-Powers model to a multi-period model
of risk-exchange among insureds, primary insurers and reinsurers and solve
for equilibrium price and quantities of insurance and reinsurance. Although
this is the main focus of our research in this chapter we also consider and
analyze different models of equilibrium pricing in insurance and reinsurance.
Chapter 5 187

5.2 Comparative analysis of different models

of insurance equilibrium pricing

5.2.1 Economics of risk sharing in insurance-reinsurance

market

Consider a market consisting of M agents (insurers and reinsurers). Each

i-th agent has its own utility function ui for which u0 > 0 and u00 0. All
agents rational behaviour is defined by maximizing expected utility. Each
agent holds a portfolio with a random payoff Xi with finite first two moments
and that belongs to a Hilbert space L2 defined on a triplet (, F, P), where

F = (X1 , X2 , ..., XM ) is a sigma-algebra generated by events associated
with random payoffs Xi . We define the market price [X] of any portfolio
with a random payoff X L2 as a functional on L2 .
It can be shown simply that unless the functional is linear, arbitrage
would be possible (e.g. see arguments on p.3 in Sherris (2002 [135])). More-
over [X] = 0 iff X = 0 a.s., otherwise an arbitrage opportunity is present.
Hence the absent of arbitrage implies that is a linear positive functional
with [0] 0, or equivalently is a continuous positive functional defined
on a Hilbert space L2 . Therefore, according to the Riesz representation
theorem (fundamental theorem of functional analysis) there exists a unique
random variable L2 such that

(X) = E[X], X L2 .

The random variable is called the state price deflator. The problem each
agent intends to solve is the following

sup E [ui (Yi )] , (5.2.1)

Yi L2
subject to [Yi ] [Xi ].
Chapter 5 188

We define a competitive equilibrium as a collection {Y1 , Y2 , ..., YM ; } such

that for each i-th agent Yi solves the optimization problem (5.2.1) and
PM P
M
Yi = Xi .
i=1 i=1
The following result about characteristic of competitive equilibrium holds
for agents with strictly monotonic utility function.

Proposition 5.2.1. The competitive equilibrium is characterized by the ex-

istence of positive constants ki , such that for the equilibrium risk sharing
X = (X1 , X2 , ..., XM )

u0i (Xi ) = ki , a.s. i = 1, ..., M

The proof of this theorem can be found in Aase (2002 [1]) and is built
on the Kuhn-Tucker theorem for functionals.

Pareto optimality

A risk exchange Y = (Y1 , Y2 , ..., YM ) is said to be Pareto optimal, if it is not

possible to make one agent better off without making the other agent worse
off. That is, there is no other risk exchange Z = (Z1 , Z2 , ..., ZM ) with

E[ui (Zi )] E[ui (Yi )], i = 1, ..., M,

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

The next most fundamental characterization of Pareto optimality is Borchs

Theorem.

Theorem 5.2.2 (Borchs Theorem). A Pareto optimal risk sharing Y is

characterized by existence of positive weights 1 > 0, 2 > 0, ..., M > 0 such
that

1 u01 (Y1 ) = 2 u02 (Y2 ) = ... = M u0M (YM ), a.s.

An elegant proof of this theorem can be found in Gerber and Pafumi

1
(1998 [66]). It follows immediately from Borchs theorem that for i = ki

we obtain i : i u0i (Yi ) = , i.e. any competitive equilibrium is Pareto

optimal. Now, taking into account the fact that the state price deflator must
1
have1 E[] = 1 we conclude that ki = E[u0i (Yi )]
.
Next we present a welfare function
M
X

E [u (W )] = sup i E [ui (Zi )] ,
P
M
Zi L2 : Zi W i=1
i=1

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

The following result represents the relationship between risk tolerance of

1 u0 (W ) 1 u0 (Y (W ))
the market r (W )
= u00 (W ) and risk tolerances ri (Yi (W ))
= u00i (Yii (W )) of all
i

agents of the market with Pareto risk sharing Y (W ) = (Y1 (W ), ..., YM (W )).

Proposition 5.2.3. The following statements hold:

1) for given deterministic value of aggregate wealth W = w the Pareto opti-
mal risk sharing Y (w) satisfies the differential equation

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

non-linear differential equation

r1 (w1 P X + I(X))
I 0 (x) = ,
r1 (w1 P X + I(X)) + r2 (w2 + P I(X))
u00
where ri = ui0 , i = 1, 2 are the absolute risk averse functions. In other
i

words, slope of I is determined by a hyperbolic relation of absolute risk

aversion functions of the both agents with boundary conditions 0 I(X)
X. This result was initially obtained by Moffet (1979 [105]) using variational
calculus. One can notice that the full risk transfer (i.e. I(X) = X) is Pareto
optimal unless the second agent is strictly risk averse. Also, any deductible
risk transfer contracts (i.e. I is any stop-loss transformation) can not be
Pareto optimal when both agents are strictly risk averse. Indeed, if both
party are strictly averse, then 0 < I 0 (X) < X for all X. However, for any
deductible form risk transfer I we have I 0 (X) equals to 0 or 1, which implies
a contradiction.
One may notice that the risk aversion of both agents is more important
than the premium for risk transfer in analysis of Pareto optimal risk sharing.
However, the Pareto optimal contract of risk sharing does depend on the
premium. Now, to determine the premium P for the Pareto optimal contract
of risk sharing IP we can use the following equation which P must satisfy
E [u0 (W ) IP (X)]
P = ,
E [u0 (W )]
since as we showed earlier the competitive equilibrium is Pareto optimal.
Alternatively, we can use the theory of cooperative games2 to determine
the premium P . In fact the competitive equilibrium premium P must be
between the two prices P2 and P1 (P2 < P1 ). As we know from classical
actuarial literature the largest premium P2 that the first agent is willing to
pay for the risk transfer IP1 is given by

E [u1 (w1 P1 X + IP1 (X))] = E [u1 (w1 X)] ,

2
For details see, for example, Baton and Lemaire (1981 [10]) and Lemaire (2004 [97]).
Chapter 5 192

and at the same time the smallest premium P2 for which the second agent
is willing to assume the risk IP2 is given by

E [u2 (w2 + P2 IP2 (X))] = E [u2 (w2 )] .

5.2.2 Financial equilibrium models of

insurance pricing

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

also be represented by the intertemporal marginal rate of substitution

u0 (c1 )
= 0 ,
u (c0 )

where c1 and c0 are an agents consumptions at the beginning and at the

end of the period respectively; u is the agents concave utility function;
is a subjective discount factor. The discount factor is stochastic because
the consumption at the end of the period is random. Taking into account
such fact of agents rational behaviour that an economic agent values money
more if it comes sooner, and if it comes in bad times when they really
need it rather than good times when they are doing well. That is why the
discount factor and hence the marginal utility of terminal consumption
are indicators of bad times. One may notice that the equilibrium price of
considered financial security can be represented in covariance form

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

One of the most fundamental financial models of insurance pricing is the

insurance CAPM which was developed in Cummins (1990 [36]; 2000 [37])
and is based on the option pricing theory (OPT). In this model the insur-
ance company holds assets and two classes of liabilities, financial debt and
Chapter 5 194

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

Solvency models of insurance pricing

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 ,

and the terminal value of insurers liability is

(
L1 , A1 > L1
1 =
A1 , A1 L1 .

So, the fair insurance premium is

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 put option, and we can see that the premium under new economic
assumption of limited liability is less than one calculated using actuarial
approach.
Under the fair (equilibrium) pricing the value of equity is

E0 = er EQ (A1 L1 ) 1{A1 >L1 } .

Summarizing, we conclude that the equilibrium insurance premium is a

solution to the following system of two equations

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 } .

Here the equilibrium price is a decreasing function of expected cost of

insolvency: a lower leverage or a higher insurers capitalization level (equity
capital) will reduce the probability of ruin, leading to a higher insurance
price. This result is consistent with the empirical hypothesis that policy-
holders are risk averse to insurers insolvency risk, i.e. they do not view
Chapter 5 197

insurance solely as a financial asset but purchase insurance in order to mit-

igate the risk of loss. Insureds cannot hedge most types of risk in any other
way than by purchasing insurance. Doing so, they accumulate their credit
exposure with one or few insurers and thus they are sensitive to the insurers
insolvency. Hence, the equilibrium (fair) insurance price is higher for high
quality insurance product, i.e. for the contract with lower insolvency risk.

5.3 Walrasian equilibrium in integrated

insurance-reinsurance market
In this section we extend the Lin-Powers (see Lin and Powers (1999 [98]))
dynamic equilibrium model of risk-exchange between insureds and insurers
in insurance market by adding new agents - reinsurers, and investigate the
equilibrium insurance and reinsurance prices, analyze the impact of reinsur-
ance market on equilibrium insurance price. It was already mentioned above
about the importance of considering a dynamic multi-period equilibrium in-
surance pricing. One of the important reasons is that insureds and insurers
tend to maximize their wealths under long-term planning horizon, and hence
the risk sharing between insureds and insurers will not necessarily be Pareto
optimal in a single-period, leading to a disequilibrium insurance price.

5.3.1 Demand for insurance

Dynamic multi-period models of insureds demand for insurance have been

developed by Briys (1986 [22]) and Gollier (1994 [68]). In their studies an in-
dividual chooses its consumption rate and insurance coverage simultaneously
to maximize utility of consumption throughout the individuals lifetime. In
Gollier (1994 [68]) for instance, the decisions are made subject to the Poisson
distribution of the insureds loss, in Lin and Powers (1999 [98]) the authors
Chapter 5 198

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

dWt = (rW Wt Ct (1 t )pt Yt t Yt ) dt + t Yt Y dBtY + Wt W dBtW ,

where

Wt denotes the insureds cumulative wealth up to time t;

Ct denotes the insureds consumption rate at time t;

Yt denotes the insureds expected incurred loss rate at time t;

pt denotes the unit price of insurance at time t;

t denotes the insureds retention level at time t;

rW denotes the constant expected rate of return on the insureds

wealth;

Y and W denote the diffusion coefficients associated with the respec-

tive diffusion processes of the insureds loss and investment return;

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

where u denotes the insureds utility function, and r is a subjective discount

rate.
Using standard methods of stochastic control theory discussed in the
previous chapter, we can write down the HJB partial differential equation

1 2 2 2 2 2
2V V V rt
max Y + Wt W + Wt + + e u(Ct ) = 0,
Ct , t 2 t t Y W 2 W t
V (T, WT ) = u(B(T, WT )).

So the insureds objective is to choose both the optimal consumption rate

Ct 0 and retention rate t [0, 1] at each instant time t. The latter PDE
is difficult to solve for general forms of the utility function u. However, in
the case of isoelastic utility with parameter < 1, it can be solved using
techniques similar to those of Merton (1992 [102], Ch.4). In this case the
optimal retention level is given by

(pt 1)Wt
t = ,
(1 ) Y2 Yt

which is

positively related to the insurance price and the insureds current

wealth;

negatively related to the expected loss rate, loss diffusion coefficient

and the insureds relative risk aversion coefficient 1 .

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.

5.3.2 The insurers supply of insurance and demand

for reinsurance

We model insurers surplus with a geometric Brownian motion

dSt = St dt + St dWt lt dt, (5.3.1)

S0 = x

where = (at , t ) = at t (L (1 P ) (L + P ) + rI pre (t)(1 at )) +

p
rI , = (a, ) = a2 2 L2 + (at + 1)2 I2 , W is a standard Brownian mo-
tion associated with the surplus, and lt is the dividend rate. The quantities
a, , L , P , rI , L , I were defined earlier in Chapter 3 and are repeated
here for convenience

a - insurers retention level of quota share proportional reinsurance;

pre - reinsurance premium rate;

p = 1 + L - gross insurance premium;

- leverage ratio;

L , P - the expense ratios for administrative expenses (proportional

to losses) and for acquisition expenses (proportional to premiums) re-
spectively;

rI , I - drift and diffusion of log-return of stock market instrument

(the companys assets are instantaneously invested in stock market
instruments);

L - diffusion of loss reserve.

Chapter 5 201

Let the regulators boundary for the insurers surplus be x < S0 = x,

so that the insurer will be declared insolvent if the surplus falls below this
boundary. Let there also be an insolvency cost K. 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 : St = x } the residual surplus net of insolvency cost
is distributed among shareholders as terminal dividends.
We assume that an insurer is risk averse with isoelastic utility function
x1m
U (x) = 1m
with m = |ln | > 0. As was shown in Chapter 4 of this thesis
the maximization of the insurers expected utility of its future dividends
is approximately equivalent to the constrained maximization of expected
present value of dividend cash flow subject to the level of probability of
ruin. Now we define the value function (shareholders value) by

Z
V (x) = V (x; a, , l, p, pre ) = sup E ers U (ls ) ds + er U (K)
, l,a
0

with boundary condition V (x ) = U (K). To achieve an optimal value V the

insurer has to find an optimal leverage ratio and dividend policy.
The corresponding HJB equation which the value function must satisfy
is

1 2 2 00
x V (x) + ( x l) V 0 (x) r V (x) + U (l) = 0, x x
2

subject to the boundary condition V (x ) = U (K). Here, and are func-

tions of , a, p, pre .
It is difficult to solve the latter HJB in the case of general form of divi-
dend payment, however as was shown in Chapter 4 it is possible to find an
analytical form of optimal value function V in the case where the dividend
payments are proportional to the surplus with proportionate coefficient
(see Section 4.4 of Chapter 4 of this thesis).
Chapter 5 202

In this model of demand for reinsurance and supply of insurance we

recognize as an index of insurance supply, and a as a retention level of
underwriting risk.

5.3.3 Supply of reinsurance

We consider a reinsurer who assumes part of the insurers underwriting risk.

This economic agent behaves like an insurer in the duopoly insured-insurer,
and thus we admit that the model of its surplus S re and objective function
are similar to that of the insurer with 1 + L = pre and a = 1. The main
reinsurers objective then is to find an optimal leverage ratio re to achieve
maximum value of expected present value of its future dividend payments
with given market reinsurance price.

5.3.4 Dynamic equilibrium insurance and reinsurance

pricing. Numerical solutions

We consider an integrated insurance-reinsurance market consisting of NP

homogeneous insureds (policyholders), NI homogeneous insurers and NR
homogeneous reinsurers. In this market the demand for insurance of each
insured is equal to (1 t )Yt , where the optimal retention is defined for
a given insurance price p. On the other hand the individual insurers supply
of insurance is equal to t St , and its demand for reinsurance is equal to
(1at )t St , where the optimal leverage ratio and the optimal reinsurance
retention ratio a are functions of given insurance price p and reinsurance
price pre . Analogously, the individual reinsurers supply of reinsurance is
re
equal to re
t St . Now, the equilibrium prices of insurance and reinsurance,
p and pre respectively, must satisfy the following system of two equations
Chapter 5 203

of aggregate insurance and reinsurance demand-supply

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

This system can be solved numerically using a discretized Markov approxi-

mation approach (or finite difference method, e.g. see Kushner and Dupuis
(1992 [96])) for computing insurance and reinsurance demand-supply curves.

Numerical example

We assume that the market consists of NP = 18 000 000 of homogeneous in-

sureds, NI = 60 homogeneous insurers and NR = 7 homogeneous reinsurers,
and consider at time t = 0

an insured, a representative of the group of NP homogeneous in-

sureds, with power (isoelastic) utility with parameter = 0.5, initial
wealth W0 = 200 000, planning horizon T = 30 years, bequest amount
B(T, WT ) = 25000, expected losses Y0 = 6000, pure insurance risk
Y2 = 15, portfolio risk W
2
= 0.2 and interest rates r = 0.05 and
rW = 0.15.

an insurer, a representative of the group of NI homogeneous insurers,

with power (isoelastic) utility with parameter m = 2.996 (which corre-
sponds to the 5% level of the insurers ruin probability), initial surplus
S0 = 300 106 , insolvency boundary x = 50 106 , insolvency cost
K = 40 106 , underwriting risk L2 = 0.6, investment risk I2 = 0.01,
interest rates r = 0.05 and rI = 0.2, expense ratios L = 0.1 and
I = 0.15.

a reinsurer, a representative of the group of NR homogeneous rein-

surers, with power (isoelastic) utility with parameter m = 4.61 (it
corresponds to the 1% level of the reinsurers ruin probability), initial
Chapter 5 204

surplus S0re = 1.2109 , insolvency boundary x = 500106 , insolvency

cost K re = 350 106 , underwriting risk (Lre )2 = 0.8, investment risk
(Ire )2 = 0.01, interest rates r = 0.05 and rI re = 0.2, expense ratios
L re = 0.1 and I re = 0.15.

The equilibrium system (5.3.2) now can be reduced to

1t (p )
t (p , pre )
0.167
(1a (p , pre )) t (p , pre )
(5.3.3)
re
(pre )
0.467
t

Solving (5.3.3) we obtain the insurance demand curve as a decreasing func-

tion of insurance market price p and the insurance supply curves as increas-
ing functions of insurance market price p under different market reinsurance
price pre . Under each reinsurance price pre it is possible to determine a
partial equilibrium insurance price as a function of reinsurance price.
Analogously to determine a partial equilibrium reinsurance price we have
to compute reinsurance demand curves as decreasing functions of reinsurance
price under different insurance prices and the reinsurance supply curve as
an increasing function of reinsurance price.
By varying the market insurance and reinsurance prices we change the
partial equilibrium prices of insurance and reinsurance until we get an equi-
librium insurance and reinsurance prices. In this numerical example the
vector (p 1.154; pre 1.217) is the solution of system (5.3.3), and also

= 0.682, = 1.908, a = 0.691, re
t = 1.263.
If we tried to solve our system under the assumption of absence of a
reinsurance market then we would get an equilibrium price of p 1.157.
Also, an increase in number of reinsurers from 7 to 10 would slightly increase
both the equilibrium insurance and reinsurance prices to approximately p =
1.167 and pre = 1.221. This effect can be explained in the following way.
In the presence of a reinsurance market the insurers are more financially
stable since they have opportunities to hedge their underwriting risk through
Chapter 5 205

reinsurance. This implies a higher quality of the insurance products supplied

by insurers, and obviously insureds are thus willing to pay higher price for
such products. Alternatively, in the absence of a reinsurance market the
insurers underwriting risk is less diversified, the insurers are less financially
stable and thus the price for insurance they provide is lower3 .

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

also considered endogenous reinsurance optimization constrained by restric-

tions on risk, which can be measured by risk measures other than variance.
For this problems we showed that under some effective risk measures of re-
tained underwriting risk an optimal reinsurance contract will be different
from change-loss reinsurance. The results of this part are very important
and gives an insight into the use of the new forms of reinsurance in practice.
The second part considered the problems of creating shareholders value
through reinsurance optimization. We have investigated the existence of
risk management incentives in shareholders value creation in the presence of
such frictional costs as corporate tax and financial distress costs. We started
from an analysis of 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 insolvency 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 to reduce its risk capital to the level under
which the minimum solvency requirements are satisfied. We showed that
there is no demand for reinsurance in the first, more conservative model
without frictional costs. However, in the presence of frictional costs such
as corporate tax and financial distress costs, 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 and costs of financial
distress. The demand for reinsurance in the second model under the absence
of frictional costs is likely due to the assumption of an actuarial premium
Chapter 6 209

principle, 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 the models with financial distress costs. We considered
modelling insurer solvency in the presence of four different forms of financial
distress costs, and in the case of Risk Based Capital 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 investment risks through reinsurance
and investment hedge. We showed that these insurance risk management
incentives increase with an increase of financial distress costs.
In the third part we analyzed the models of dynamic risk management in
the maximization of shareholders value, defined as expected present value
of future dividend payments. Here the reinsurance and dividend payments
are the main control variables of insurer risk management. This part starts
from the analysis of different stochastic control models of dynamic insurance
risk management, namely models in which the expected value of future, un-
til ruin time, discounted dividend payments are controlled via two control
variables: dividend and reinsurance policies. We showed that these models
impose 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 of the risk-free rate of return is decreasing func-
tion 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 companys surplus with dividend
Chapter 6 210

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

[1] Aase, K. Perspectives of risk sharing. Scandinavian Actuarial Jour-

nal, Vol. 2:pp. 73128, 2002.

[2] Andrade, G. and Kaplan, S. How costly is financial (not eco-

nomic) distress? Evidence from highly levereged transactions that
became distressed. Journal of Finance, Vol. 53:pp. 14431493, 1998.

[3] Arrow, J. Essays in the Theory of Risk Bearing. North Holland,

Chicago, 1971.

[4] Arrow, K.J. Le role des valeurs boursieres pour la repartition la

meilleure des risques. Econometrie, Paris: CNRS:pp. 4147, December
1953.

[5] Arrow, K.J. Uncertainty and the welfare economics of medical care.
American Economic Review, Vol. LIII:pp. 941973, December 1963.

[6] Arrow, K.J. and Debreu, G. Existence of equilibrium for a com-

petitive economy. Econometrica, Vol. 22:pp. 265290, 1954.

[7] Asmussen, S. Ruin probabilities. World Scientific, Singapour, 2000.

[8] Asmussen, S. and Taksar, M. Controlled diffusion models for

optimal dividend pay-out. Insurance: Mathematics and Economics,
Vol. 20:pp. 115, 1997.

212
Bibliography 213

[9] Asmussen, S., Hjgaard, B. and Taksar, M. Optimal risk

control and dividend distribution policies. example of excess-of-loss
reinsurance for an insurance corporation. Finance and Stochastics,
Vol. 4:pp. 299324, 2000.

[10] Baton, B. and Lemaire, J. The core of a reinsurance market.

ASTIN Bulletin, Vol. 12:pp. 5771, 1981.

[11] Berezansky, Y., Sheftel, Z. and Us, G. Functional Analysis.

Birkhauser Verlag, 1996.

[12] Biger, N. and Kahane, Y. Risk considerations in insurance

ratemaking. The Journal of Risk and Insurance, Vol. 44(1):pp. 121
132, 1978.

[13] Blazenko, G. Optimal insurance policies. Insurance: Mathematics

and Economics, Vol. 4, No 4:pp. 267278, 1985.

[14] Borch, K. An attempt to determine the optimum amount of stop-

loss reinsurance. Transactions of the XVI International Congress of
Actuaries, Vol. 1:pp. 597610, 1960.

[15] Borch, K. Reciprocal reinsurance treaties. The ASTIN Bulletin,

Vol. 1, part IV:pp. 170191, 1960.

[16] Borch, K. Equilibrium in a reinsurance market. Econometrica, Vol.

30:pp. 424444, 1962.

[17] Borch, K. The mathematical theory of insurance. Lexington, MA:

Lexington Books, 1974.

[18] Borch, K. A theory of insurance premiums. The Geneva Papers on

Risk and Insurance, Vol. 10:pp. 192208, 1985.
Bibliography 214

[19] Borch, K. Economics of insurance. North-Holland, Amsterdam,

1990.

[20] Bowers, N., Gerber, H., Hickman, J., Jones, D. and Nes-
bitt, C. Actuarial mathematics. Society of Actuaries, 2nd ed.
Schaumburg, Ill., 1997.

[21] Brealey, R. and Myers, S. Principle of corporate finance.

McGraw-Hill, 1999.

[22] Briys, E. Insurance and consumption: the continuous-time case.

Journal of Risk and Insurance, Vol. 53:pp. 718723, 1986.

[23] Briys, E. and de Varenne, F. Valuing risky fixed rate debt: an

extension. Journal of Financial and Quantitative Analysis, Vol. 32,
No 2 (June):pp. 239248, 1997.

[24] Brockett, P and Xia, X. Operations research in insurance: a

review. Transactions of the Society of Actuaries, Vol. XLVII:pp. 780,
1995.

[25] Buhlmann, H. Mathematical methods in risk theory. Springer, New

York, 1970.

[26] Buhlmann, H. An economic premium principle. ASTIN Bulletin,

Vol. 11:pp. 5260, 1980.

[27] Chen, R. and Venezian, E. Insurance pricing in a continuous-time,

multi-period framework. Proceeding of the 4th International Confer-
ence of Insurance Solvency and Finance, 1994.

[28] Choulli, T., Taksar, M. and Zhou, X. A diffusion model for

optimal dividend distribution for a company with constraints on risk
control. SIAM Journal of Control Optimization, Vol. 41, No 6:pp.
19481979, 2003.
Bibliography 215

[29] Cochrane, J. Asset pricing. Princeton University Press, 2001.

[30] Cologne Re. Risk Insights Technical Publications.

http://www.colognere.com, 1999-2004.

[31] Converium Re. Technical Publications. http://www.converium.com,

1999-2004.

[32] Cramer, H. On the mathematical theory of risk. Skand. Jubilee

Volume, Stockholm, 1930.

[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.

[35] Cummins, D. Risk-based premiums for insurance guaranty funds.

Journal of Finance, Vol. 43:pp. 823839, 1988.

[36] Cummins, D. Asset pricing models and insurance ratemaking. ASTIN

Bulletin, Vol. 20:pp. 125166, 1990.

[37] Cummins, D. Applications of financial pricing models in property-

liability insurance. In G. Dionne, editor, Handbook of Insurance.
Kluwer Academic Publishers, Boston, 2000.

[38] Cummins, D. and Sommer, D. Capital and risk in property-liability

insurance markets. Journal of Banking and Finance, Vol. 20:pp. 1069
1092, 1996.

[39] Dassios, A. and Embrechts, P. Martingales and insurance risk.

Comm. Stat. Stoch. Models, Vol. 5:pp. 181217, 1989.

[40] Davis, M.H.A. Markov models and optimization. Chapman & Hall,
London, 1993.
Bibliography 216

[41] Daykin, C., Pentikanen, T. and Pesonen, M. Practical risk

theory for actuaries. Chapmen & Hall, New York, 1996.

[42] Debreu, G. Une Economie de lincertain. Mimeo, Electricite de

France, 1953.

[43] de Finetti, B. Il problema dei pieni. Giorn. Ist. Ital. Attuari, Vol.
11:pp. 188, 1940.

[44] de Finetti, B. Su unimpostacione alternativa dell teoria collectiva

del rischio. Transaction of the XV International Congress of Actuaries,
New York, Vol. 2:pp. 433443, 1957.

[45] Deprez, O. and Gerber, H. On convex principles of premium

calculation. Insurance: Mathematics and Economics, Vol. 4:pp. 179
189, 1985.

[46] Diamond, D.W. Debt maturity structure and liquidity risk. Quaterly
Journal of Econimics, Vol. 106:pp. 709737, 1991.

[47] Dickson, D. and Waters, H. Reinsurance and Ruin. Insurance:

Mathematics and Economics, Vol. 19:pp. 6180, 1996.

[48] Doherty, N. Integrated risk management: techniques and strategies

for managing corporate risk. McGraw-Hill, New York, 2000.

[49] Doherty, N. and Garven, J. Price regulation in property-liability

insurance: A contingent claims approach. Journal of Finance, Vol.
41:pp. 10311050, 1986.

[50] Duffie, D. Dynamic asset pricing theory. Princeton University Press,

Princeton, 2001.

[51] DuMouchel, W. The pareto optimality of an n-company reinsurance

treaty. Scandinavian Actuarial Journal, Vol. 51:pp. 165170, 1968.
Bibliography 217

[52] Ehrlich, I. and Becker, G. Market insurance, self-insurance, and

self-protection. Journal of Political Economy, Vol. 80:pp. 623648,
1972.

[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.

[54] Etheridge, A. Financial Calculus. Cambridge UP, Cambridge, 2002.

[55] Fairley, W. Investment income and profit margins in property-

liability insurance: Theory and empirical results. Bell Journal of Eco-
nomics, Vol. 10:pp. 192210, 1979.

[56] Fleming, W.H. and Rishel, R.W. Deterministic and stochastic

optimal control. Springer, New York, 1975.

[57] Fleming, W.H. and Soner, H.M. Controlled Markov processes

and viscosity solution. Springer-Verlag, New York, 1993.

[58] Froot, K., Scharfstein, D. and Stein, J. Risk management:

coordinating corporate investments and financing policies. Journal of
Finance, Vol. 5:pp. 16291658, 1993.

[59] Gajek, L. and Zagrodny, D. Insurers optimal reinsurance strate-

gies. Insurance: Mathematics and Economics, Vol. 27:pp. 105112,
2000.

[60] Gajek, L. and Zagrodny, D. Optimal reinsurance under general

risk measures. Insurance: Mathematics and Economics, Vol. 34:pp.
177390, 2004.

[61] Garven, J. On the application of finance theory to the insurance

firm. Journal of Financial Services Research, Vol. 1:pp. 5776, 1987.
Bibliography 218

[62] Geman, H., El Karoui and Rochet, C. Change of numeraire,

changes of probability measure and option pricing. J.Appl.Prob., Vol.
32:pp. 443458, 1995.

[63] Gerber, H. Games of economic survival with discrete- and

continuous-income processes. Oper. Res., Vol. 20:pp. 3745, 1972.

[64] Gerber, H. An introduction to mathematical risk theory. S.S. Hueb-

ner Foundation, Philadelphia, PA, 1979.

[65] Gerber, H. The esscher premium principle: A criticism. The ASTIN

Bulletin, Vol. 12:pp. 139140, 1981.

[66] Gerber, H. and Pafumi, G. Utility functions: from risk theory to

finance. North American Actuarial Journal, Vol. 2, No 3:pp. 74100,
1998.

[67] Gerber, H. and Shiu, E. Optimal dividends: analysis with brow-

nian motion. North American Actuarial Journal, Vol. 8, No 1, 2004.

[68] Gollier, C. Insurance and precautionary capital accumulation in a

continuous-time model. Journal of Risk and Insurance, Vol. 61(1):pp.
7895, 1994.

[69] Gollier, C. Optimal insurance design: What can we do with and

without expected utility. In G. Dionne, editor, Handbook of Insurance.
Kluwer Academic Publishers, Boston, 2000.

[70] Goovaerts, M., De Vylder, F. and Haezendonck, J. Insur-

ance Premims. North-Holland, Amsterdam, 1984.

[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

[72] Grandell, J. A class of approximations of ruin probabilities. Scan-

dinavian Actuarial Journal, suppl.:pp. 3752, 1977.

[73] Grandell, J. A remark on a class of approximations of ruin prob-

abilities. Scandinavian Actuarial Journal, No 2:pp. 7778, 1978.

[74] Grandell, J. Aspects of risk theory. Springer-Verlag, New York,

1990.

[75] Grundl, H. and Schmeiser, H. Pricing double-trigger reinsurance

contracts: financial versus actuarial approach. Journal of Risk and
Insurance, Vol. 69, No 4:pp. 449468, 2002.

[76] Hipp, C. . Stochastic control with application in insurance. Technical

Report, University of Karslruhe, Germany, 2003.

[77] Hipp, C. and Plum, M. Optimal investment for insurers. Insurance:

Mathematics and Economics, Vol. 27:pp. 215228, 2000.

[78] Hipp, C. and Plum, M. Optimal investment for investors with

state dependent income, and for insurers. Finance and Stochastics,
Vol. 7:pp. 299321, 2003.

[79] Hipp, C. and Schmidli, H. Asymptotics of the ruin probability

for the controlled risk process: the small claims case. Scandinavian
Actuarial Journal, Vol. 7:pp. 299321, 2003.

[80] Hipp, C. and Taksar, M. Stochastic control for optimal new busi-
ness. Insurance: Mathematics and Economics, Vol. 26:pp. 185192,
2000.

[81] Hipp, C. and Vogt, M. Optimal dynamic XL reinsurance. ASTIN

Bulletin, Vol. 33:pp. 193208, 2003.
Bibliography 220

[82] Hjgaard, B. and Taksar, M. Controlling risk exposure and div-

idends pay-out schemes: Insurance company example. Math.Finance,
Vol. 2:pp. 153182, 1999.

[83] Hjgaard, B. and Taksar, M. Optimal risk control for a large

corporation in the presence of returns on investments. Finance and
Stochastics, Vol. 5:pp. 527547, 2001.

[84] Hurlimann, W. Solvabilite et reassurance. Bulletin of the Swiss

Association of Actuaries, pages pp. 229249, 1993.

[85] Iglehart, D.L. Diffusion approximation in collective risk theory.

J.Appl.Probab., Vol. 6:pp. 285292, 1969.

[86] Ioffe, A. and Tikhomirov, V. Theory of Extremal Problems.

Nauka, Moscow (in Russian), 1974.

[87] Jacod, J. and Shiryev, A. Limit theorems for stochastic processes.

Springer, Berlin, 1988.

[88] Jarrow, R. and Purnanandam, A. Capital structure and the

present value of firms investment opportunities: a reduced form credit
risk perespective. Working Paper, Johnson Graduate School of Man-
agement, Cornell University, NY, January 15, 2004.

[89] Jeanblanc-Picque, M. and Shiryaev, A.N. Optimization of the

flow of dividends. Russian Mathematical Surveys, Vol. 50:pp. 257277,
1995.

[90] Kaas, R., Goovaerts, M., Dhaene, J. and Denuit, M. Modern

actuarial risk theory. Kluwer AP, Boston, 2001.

[91] Kahn, P. Some remarks on a recent paper by borch. The ASTIN

Bulletin, Vol. 1:pp. 265272, 1961.
Bibliography 221

[92] Kaluszka, M. Optimal reinsurance under mean-variance premium

principles. Insurance: Mathematics and Economics, Vol. 28:pp. 6167,
2001.

[93] Karatzas, I. and Shreve, S. Brownian motion and stochastic

calculus. Springer, New York, 1988.

[94] Krvavych, Y. and Sherris, M. Enhancing insurer value through

reinsurance optimization. UNSW Working Paper, July 2004.

[95] Krylov, N.V. Controlled diffusion processes. Springer-Verlag, New

York, 1980.

[96] Kushner, H. and Dupuis, P. Numerical methods for stochastic

control problems in continuous time. Springer-Verlag, 1992.

[97] Lemaire, J. Borchs theorem. In J. Teugels and B. Sundt, editors,

Encyclopedia of Actuarial Science. Wiley, 2004.

[98] Lin, W. and Powers, M. Insurance market equilibrium: a

multi-period dynamic solution. Technical Report, Temple University,
Philadelphia, USA, 1999.

[99] Lundberg, O. Uber die Theorie der Ruckversicherung. Transac-

tions of the VI International Congress of Actuaries, Vol. 1:pp. 877
948, 1909.

[100] Markowitz, H. Portfolio selection: efficient diversification of in-

vestments. Wiley, Yale University Press, 1959.

[101] Martin-Lof, A. Lectures on the use of control theory in insurance.

Scandinavian Actuarial Journal, Vol. 77:pp. 125, 1994.

[102] Merton, R. Continuous-time finance. Blackwell, 1992.

Bibliography 222

[103] Merton, R. and Perold, A. Theory of risk capital in financial

firms. In J.M. Stern and D.H. Chew, editors, The revolution in corpo-
rate finance. Malden, MA: Blackwell Business, Malden, MA, 1998.

[104] Miller, M.H. and Modigliani, G. Dividend policy, growth and

valuation of shares. Journal of Business, Vol. 34:pp. 411433, 1961.

[105] Moffet, D. The risk sharing problem. Geneva Papers on Risk and
Insurance, Vol. 11:pp. 513, 1979.

[106] Morill, J. One-person games of economic survival. Naval Research

Logistic Quaterly, Vol. 13:pp. 4970, 1966.

[107] Mossin, J. Aspects of rational insurance purchasing. Journal of

Political Economy, Vol. 79:pp. 553568, 1968.

[108] Musiela, M. and Rutkowski, M. Martingale methods in financial

modelling. Springer, 1998.

[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.

[110] Nakada, P., Shan, H., Koyluoglu, H. and Collingon, O.

P& c raroc: A catalyst for improved capital management in the prop-
erty and casualty insurance industry. THE JOURNAL OF RISK FI-
NANCE, 1999.

[111] Ohlin, J. A generalization of a result by Borch and Kahn on the

optimal properties of stop-loss reinsurance. The ASTIN Bulletin, Vol.
5:pp. 249266, 1969.

[112] ksendal, B. Stochastic differential equations. Springer-Verlag, New

York, 2000.
Bibliography 223

[113] Opler, T. and Titman, S. Financial distress and coporate perfor-

mance. Journal of Finance, Vol. 49:pp. 10151040, 1994.

[114] Outreville, F. Theory and Prctice of Insurance. Kluwer AP, 1998.

[115] Panjer, H. et al. Financial economics. The Actuarial Foundation,

Schaumburg, Ill, 1998.

[116] Paulsen, J. Optimal dividend payouts for diffusions with solvency

constraints. Finance and Stochastics, Vol. 7:pp. 457473, 2003.

[117] Paulsen, J. and Gjessing, H. Optimal choice of dividend barriers

for a risk process with stochastic return on investment. Insurance:
Mathematics and Economics, Vol. 20:pp. 215223, 1997.

[118] Peressini, A., Sullivan, F. and Uhl, J. The Mathematics of

Nonlinear Programming. Springer-Verlag, New York, 1988.

[119] Polyanin, A. and Zaitzev, V. Handbook of exact solutions for

ordinary differential equations. Chapman & Hall/CRC, 2nd edition,
2002.

[120] Polyanin, A. and Zaitzev, V. Handbook of nonlinear partial dif-

ferential equations. Chapman & Hall/CRC, 2003.

[121] Powers, M. A theory of risk, return and solvency. Insurance: Math-

ematics and Economics, Vol. 17:pp. 101118, 1995.

[122] Rajani, S. Insolvency costs and fees. Tolley Publishing Co, 2002.

[123] Raviv, A. The design of an optimal insurance policy. American

Economic Review, Vol. 69:pp. 8496, 1979.

[124] Rolski, T., Schmidli, H., Schmidt, V. and Teugels, J.

Stochastic Processes for Insurance and Finance. Wiley, Chichester,
1999.
Bibliography 224

[125] Schal, M. On piecewise deterministic markov control process: con-

trol of jumps and of risk processes in insurance. Insurance: Mathe-
matics and Economics, Vol. 22:pp. 7591, 1998.

[126] Schlesinger, H. The theory of insurance demand. In G. Dionne,

editor, Handbook of Insurance. Kluwer Academic Publishers, Boston,
2000.

[127] Schmidli, H. Diffusion approximation for a risk process with posi-

ibility of borrowing and interest. Stochastic Models, Vol. 10, 1993.

[128] Schmidli, H. Optimal proportinal reinsurance policies in a dynamic

setting. Scandinavian Actuarial Journal, Vol. 1:pp. 5568, 2001.

[129] Schmidli, H. On minimising the ruin probability by investment and

reinsurance. Ann. Appl. Probab., Vol. 12:pp. 890907, 2002.

[130] Schonbucher, P. Credit derivatives pricing models: models, pricing

and implimentation. Wiley, 2003.

[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.

[132] Sethi, S.P., Derzko, N.A. and Lehoczky, J. General solution

of the stochastic price-dividend integral equation: a theory of financial
valuation. SIAM Journal of Mathematical Analysis, Vol. 15:pp. 1100
1113, 1984.

[133] Sethi, S.P., Derzko, N.A. and Lehoczky, J. A stochastic ex-

tention of miller-modigliani framework. Mathematical Finance, Vol.
1(4):pp. 5776, 1991.

[134] Sherris, M. Solvency, capital allocation and fair rate of return in

insurance. UNSW Working Paper, January 2004.
Bibliography 225

[135] Sherris, M. Economic valuation: something old, something new.

UNSW Working Paper, October 2002.

[136] Sherris, M. and van der Hoek, J. Capital allocation in insurance:

Economic capital and the allocation of the default option value. UNSW
Working Paper, August 2004.

[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.

[138] Smith, V. Optimal insurance coverage. Journal of Political Economy,

Vol. 76:pp. 6877, 1968.

[139] SOA Economic Capital Calculation and Allocation Sub-

group. Specialty guide on economic capital. Technical Report, June
2003.

[140] Spaeter, S. and Rogers, P. Administrative costs and optimal

insurance contracts. LARGE Preprint No10, Universite Louis Pasteur,
Strasbourg, 1995.

[141] Swiss Re. Technical Publications. http://www.swissre.com, 1999-

2004.

[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.

[144] Taksar, M. and Markussen, C. Optimal dynamic reinsurance

policies for large insurance portfolios. Finance and Stochastics, Vol.
7:pp. 97121, 2003.
Bibliography 226

[145] Varian, H. Microeconomic analysis. Norton, 1992.

[146] Venezian, E. Some thoughts on the relation between the supply of

and demand for insurance. Proceeding of 1994 Annual Meeting of the
ARIA, 1994.

[147] Venter, G. Premium calculation implications of reinsurance without

arbitrage. ASTIN Bulletin, Vol. 21, No 2:pp. 223256, 1991.

[148] Waters, H. Some mathematical aspects of reinsurance. Insurance:

Mathematics and Economics, Vol. 2:pp. 1726, 1983.