Upload
chicouiobi
View
217
Download
0
Embed Size (px)
Citation preview
8/9/2019 252028.pdf
1/24
Risk Management and the Theory of the FirmAuthor(s): J. David Cummins
Source:The Journal of Risk and Insurance,
Vol. 43, No. 4 (Dec., 1976), pp. 587-609Published by: American Risk and Insurance AssociationStable URL: http://www.jstor.org/stable/252028 .
Accessed: 11/06/2014 17:25
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp
.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].
.
American Risk and Insurance Association is collaborating with JSTOR to digitize, preserve and extend access
to The Journal of Risk and Insurance.
http://www.jstor.org
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/action/showPublisher?publisherCode=arihttp://www.jstor.org/stable/252028?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/252028?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=ari
8/9/2019 252028.pdf
2/24
Risk
Management
and
the
Theory
of the
Firm
J.
DAVID CUMMINS
ABSTRACT
This
article integrates risk
management
decision variables into the
theory
of the
firm under
risk
and
develops
risk
management
decision
rules consistent with
the firm's overall
objectives.
The
theoretical con-
struct chosen for extension to the risk management problem is the
capital asset
pricing model.
Utilizing
this
model,
decision
rules
are
developed
for
optimal
proportional
retention, selection
of
aggregate
deductibles and
choosing
reserving policies.
The results
differ
signifi-
cantly from
the
expected
value
decision rules
developed by previous re-
searchers.
The
article
concludes by
examining
the
impact
on
the decision
rules of
relaxing
some of
the key assumptions
of
the model.
Recent
years
have
witnessed
significant
progress
in the
application of
quantitative
decision
tools to the
solution
of risk
management
problems.
These developments have ranged from purely theoretical exercises to highly
particularized practical
applications and
have
encompassed
the
entire
spectrum of
quantitative
methodology. For
example, a model
for
the
selection of
optimal
deductibles has been
developed and
applied
by Allen
and
Duvall
[1,
4].
Optimal
insurance and
loss
prevention
decisions
have
been
the
subject
of
a
paper
by
Shpilberg and
de
Neufville
[16]
which
employed a
decision
theoretic
framework. A
study which
involved the
testing of
alternative decision
tools,
such as
utility
theory and
the
worry
factor method, has been conducted by Neter and Williams [13] while
a
computer simulation
model of self
insurance
of
workers'
compensation
losses
was
developed by Mortimer
[11].
The
fitting of
loss
distributions
to
facilitate
the
more
precise estimation
of
future claim
costs
has
been em-
phasized
by Hartman
and
Siskin
[9].
Finally, Head
[8]
has
illustrated
how
capital
budgeting
can be
applied in
the context
of
risk management
decision
making.
These
and
numerous
other studies
have
provided the basis
for
significant
improvements in
risk
management
analysis.
However, as
is perhaps
in-
evitable in a developing field, most of the articles have concentrated on
J.
David
Cummins
is
Assistant
Professor
of
Insurance
in the Wharton
School, Univer-
sity
of
Pennsylvania.
Professor
Cummins is
Research Director
of the
S. S.
Huebner
Foundation and
is
a member
of
the
Editorial
Board
of the
Journalof Risk
and
Insurance.
He
is
the author
of An
Econometric
Model
of
the
Life
Insurance Sector of
the U.S.
Economy and
co-authorof
Consumer
Attitudes
Toward
Auto and
Homeowners
Insurance.
This
paper
was
submitted for
publication
June,
1975.
It was
presented to
the
1975
Risk
Theory
Seminar.
(
587)
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
3/24
588 The
Journal
of Risk and Insurance
local rather
than
global optimization.
That is, the overall objectives of
the
firm generally
have been
recognized
only implicitly or
peripherally
in
the formulation
of the
risk
management
decision
rules.' This approach
is
potentially misleading,since rules which appear appropriatestrictly from
a
risk
management
point of
view may be
suboptimal
in the context
of
the
firm's broader goals. An
appropriatecomplement
to the
existing risk
management literature
thus
would seem to be
a general
theory which
integrates
risk management
nto the overall
business setting.
The purpose
of
this article is to develop
such a theoretical
framework.
As microeconomictheory
is a highly
developed discipline,
this
paper
does not attempt
to derive a totally
new theoretical
model. Rather,
the
goal
is
to adapt existing models
to incorporate
risk management
variables
and parameters.Because of the nature of the risk managementproblem,
a theoretical model is required
which
views the firm in
the context
of
uncertainty.
After
consideringa number
of alternatives, he
model chosen
for scrutiny in this paper
is the capital asset
pricing model
developed
by
finance theorists.2 The applicability of this
model was
suggested
to
the author by a recent
paper by Schramm
and Sherman [15].
That article
exploredthe idea
that firms
can managethe variability
of their
net income
streams
by varying
their advertising
and research and development
ex-
penditures. The
present paper
focuses on the
effect of the
selection of
alternativerisk managementdevices on that variability.In this regard,the
capital asset pricing
model
is first utilized to
derive some
fundamental
optimization
rules.
These rules
are
then expressed
in
terms
of familiar
risk
management
variables. Because
the
model rests
on
a
number of
simplifying
assump-
tions,
the paper
explores
the
realism
of
some
of
these
assumptions
n the
risk management
context and the impact
of their
relaxationon
the decision
rules. The paper
concludes
with a discussion
of the
potential
for the em-
pirical applicationof the model. The emphasisof this article is on theory
rather
than
on
practical applications
and the
analysis
is
conducted at a
high
level
of
abstraction.
Hopefully,
the theoretical
results
presented
eventually
will
provide guidelines
for
the
development
and testing
of
practical
risk management
decision
rules.
However,
much research
remains
before
direct
practical application
is
possible.
The
Capital
Asset
Pricing
Model
The
capital
asset
pricing
model as
developed
in the
theory
of
finance
is
a two period model in which the participantsare consumersand business
firms.
At
the
beginning
of Period
1,
consumersare assumed
to come to
the
market with
"quantities
of
resources-labor,
which
will
be sold to
some
firm,
and portfolioassets,
that
is,
shares
of firms
...
that must
be allocated
to
current
consumption
cl
.
.
. and
a
portfolio
investment
whose market
1
This
problem
has been recognized
in an
earlier
paper
by
Mehr and
Forbes
[10].
2This
theory
is
discussed
in
numerous
books
and articles including Fama
and
Miller
[51
and
Mossin
[11].
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
4/24
Risk
Management
589
value
at
the
beginning of Period
2
determines
the
level of
consumptions c2"
for that
period.
Consumers
are assumed to be risk averse
expected
utility
maximizers
and
to
behave as
if
the
probability
distributions
of return
on
their portfolios can be completely summarized in terms of two para-
meters,
the
expected
value
and some measure
of
the
dispersion
of the
returns.
The business
firms
in
the market organize
production
activities which
are
financed
by
selling
shares in their Period 2
output
values to
consumers
The
firms
are assumed
to operate
according to the decision rule: maximize
Period I
market value.4
The solution of the
model involves
the
simultaneous
optimal
allocation of funds by
consumers
between consumption
cl and
investment and the
maximization by
firms of their Period 1
market values.
A number of assumptions are inherent in the capital asset pricing model:
1.
Perfect
capital markets. This
assumption
is similar to that found
in the
conventional
economic theory
of perfect competition.
Among its
major
implications for the instant
case are that
no transactions costs
or taxes
exist,
and that
"no
firm is
large enough to affect
the
opportunity set
facing consumers."
2.
Consumerrisk aversion.
3.
Symmetric
stable
distributions. Consumers
are
assumed to behave as if
distributions
of returns on all
portfolios
are symmetric stable with
the
same value
of
the
characteristicexponent. This
assumption permits
the
summarization
of returndistributions
by two
parameters-the mean
and
a
measure of
dispersion.
For
convenience, this analysismakes
the added
assumption
that
the distributions are
normal;
i.e.,
that
the
measure
of
dispersion
s
the
variance.
4.
Riskless
borrowing
and
lending.
The
assumption
is
that consumers can
borrow and lend
as
much
as
they
wish at a
risk
free rate of
interest.
5.
Homogeneous
expectations.
This
assumption
means
that
"all
consumers
have
the same set
of
portfolioopportunities
. .
and all view
the
prob-
ability
distributions of
returns
associated with the various available
portfolios n the same
way."
6
All
five
major
assumptions
are
crucial
to
the
model,
but the last
four
make the
capital
asset
pricing
model
unique.
The
third
assumption
im-
plies that consumers
can summarize
all
distributions of
return in terms
of
the mean
and
a measure
of the
dispersion
of those
distributions. Com-
bined with
the
second
assumption,
the
use of
two
parameter
distributions
permits
the consumer
to consider
only
those
distributions
which are
mean,
variance
efficient. Assumptions
4
and
5 further
simplify
the
problem by
permitting
all
consumers
first to
view
the
efficient set
in
the
same
way
(assumption 5) and then (through assumption 4) to focus attention on
one
common
point
on that
frontier,
the
market
portfolio.
Thus, the
assump-
3
Fama and Miller
[5],
p.
277.
4For a
more complete
discussion
of
the firm's
objectives
see Fama
and Miller [5],
pp.
299-301. Schramm
and Sherman
[15]
have
shown
that when risk is taken
into
account,
the value maximization
goal may
give
rise
to
business behavior
which
manifests
itself
as either
growth
or
sales
maximization.
5
Fama
and
Miller
[51,
p.
277.
i
7i2A
D.
9287
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
5/24
.590
The Journal
of
Risk
and
Insurance
tions
lead
to
market
equilibrium
conditions
in
which
individual
utility
functions are
not
explicitly
present.
Although capital
market equilibrium
in
the foregoing
model
implies
a number of interesting relationships, the one selected for the analysis
of
risk
management
decisions
is
the
pricing equation
for the
Period 1
value of
a firm:
E(l)Rf
(1)
where
P
-
the
equilibrium
market
value of
the
jth
firm
at the
beginning
of
Pqriod
1
(in
equilibrium,
this
quantity
is
maximized);
- the market value of the firm at the beginning of Period 2;
E(f) -
P i
+
Rf)
S
M
. -f
m
(
-'the market value
at the
beginning
of Period 2 of the
market
m
portfolio
("market
wealth");
PM
market
value
at the
beginning
of
Period
1
of
the
market
portfolio;
Rf
a
the risk free borrowing-lending rate; and
p
a
the correlation
coefficient
between
the return on the
jth
firm
and that on the market
portfolio.
The
placement of a
tilde over
a symbol
indicates that it
represents a
random variable.
According
to the
pricing
equation,
the Period
1
marketvalue
of the
firm
is the
present
value,
computedat the
risk
free rate, of the
firm's
expected
value at Period 2 less the present value of a risk charge. The risk charge
is
equal
to the
product of
the market
price of
a unit of
standard
deviation
(Sm),
the
correlation
coefficient
of the
firm'sreturn
with
that of the
market
and
the
standard
deviation
of the firm's
Period 2 market
value. Of
interest
is
that the
risk
penalty of the
jth finn
varies
directly
with its
correlation
coefficient
with market
return.
That
relationship exists
because a stock
with
a
relatively
low correlation
with
market
return is
more valuable
to
investors in terms
of
diversification.
Optimal Risk ManagementDecisions
General
Optimization
Rules
In
most
developments
involving the
capital
asset pricing
model,
the
parameters of
the
distribution
of
returns of the
individual
firms
remain
invariant. In
order
to apply the
model
in the risk
managementcase,
how-
ever, that
assumption
must be relaxed
to
permit the firm
to
vary within
7
Ibid., p.
296.
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
6/24
Risk
Management
591
limits
its mean
and
variance
of
return.
In particular,
he assumption
that
the
objective
of
the
firm
is
to vary
E(V)
and
a(V1)
to maximize
ts
Period
1 value,
P5.
This
procedure
is an
accurateabstraction
of
the
risk
manage-
ment process as most decisions involving the use of deductibles and self
insurance
hinge
on the
tradeoff
between
savings
in the expected
cost
of
pure
risk
and
the
increased
risk
faced
by
the
firm
as a
result
of
reductions
in
its
insurance
coverage.
For
example,
when
a
firm
elects
a larger
deductible
the
usual
motivation
is
that
its insurance
premium
is reduced
by
a
larger
amount
than
the
resultant
increase
in
expected
retained
claims
and
settlement
costs.
In
return
for
this
savings,
however,
the
firm
must
be
willing
to
accept
a
higher
degree
of
variability
in
its
income
stream
as
an
unexpectedly
large
numberof losses could result in total loss costs substantially n excess of
the
original
premium.
Intuitively,
the anticipation
is
that
the
firm
would
increase
its
deductible
to
the
point
at which
the
value
of
the
premium
savings
is offset
precisely
by the
cost
of
the
increased
variability
in
the
income
stream.
A
noteworthy
assumption
in this analysis
is
that
the
firm
can
obtain
a
net
savings
by
varying
its degree
of
pure
risk retention.
Such a
savings
is
not
necessarily
available
in the world
of
perfect
markets
and
the
assump-
tion is equivalent to the introductionof a degree of imperfectionin the
insurance
market.
Investigations
of the
precise
role
of
insurance
n
economic
equilibrium
and the
conditions
under
which
the
hypothesized
savings
would
be
available
are
beyond
the
scope
of
this
study
but
constitute
in-
triguing
avenues
for
future
research.
This
risk
management
process,
under
the
assumption
that
savings
through
risk
retention
are
available,
can
be
expressed
in a
more
precise
manner
through
the use
of the capital
asset pricing
model.
Thus,
accepting
a larger
deductible
or
adopting
a
self
insurance
program
results
in
an
increase
in both
E(V' and
O);
I-e.,,
;D
>o
and
3D.L
>
0
where
D
stands
for
the
retention
of
the insured
The
optimization
rule thus
becomes:
Maximize:
Equation
(1)
with
respect
to D.
In
general
terms,
the
first-
and second-order
conditions
for a maximum
are
the following:
First-order
Condition
3D 3D
mjm
a3D
no
UD
)/(
aD
is
P
(2t
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
7/24
592
The Journal
of
Risk and
Insurance
Second-order
Condition
a2p
921( )
32a('
)
BD2 3D2
D
3D2
When written as
(2)',
the
first
order
condition for a
maximum
can
be
given
an
interesting
interpretation.
Implied
is
that the
firm
should
in-
crease its
risk
retention to the
point
at
which
the
marginal
rate
of
sub-
stitution
between return
and risk is
equal
to
the
price placed
by the
market
on a unit
of
standard deviation
multiplied by
the
correlation
co-
efficient between
the firm
and the
market.
Beyond
that
point,
the firm
will not
receive
an
increase in
expected
return
adequate
to
compensate
for
the additional
risk.
Those familiarwith the capital asset pricing model will recognize that
the
derivation of
expression (2)
involves
additional
simplifying
assump-
tions.
In
particular
the
assumption
that the risk
management
decisions
of
the
individualfirm have
no
impact either
on
the
marketprice
of
risk,
S.,
or the
correlation
between the
jth
firm's
return
and
that
of
the
market;
i.e.,
the
assumption
s
that
aSm/JD
and
apjm/aDare
equal
to zero.
Those
assumptions
are not
supportable
in
a strict
mathematical sense as
the
parameters
of
the
firm are
components of the market
parameters.
How-
ever,
the
former
condition can
be justified
by
noting that the
model
is
couched in terms of perfect competition. Thus, even though the firm's
risk-return
characteristicsare
components of
Sm,
the number of firms
in
the market is
so large
that any
change on
the part of
one
particular
firm
has no
measurable
effect
on the market
set.
The
latter
condition, ap
/1D
=
0,
is
considerably
more
restrictive, but
its
use in an
approximate
ense can be
justified
heuristically
by recognizing
that
most of the
covariancebetween
the
returns of the
various
firms
in
the
market
arises
from the
nature
of their
responseto
economic
fluctuations.
A change in retention, on the other hand, involves the assumption of
nsks
generally
uncorrelatedwith
the business
cycle
and are
approximately
independent across
the
set of all firms.8
Thus,
additional
pure risk as-
sumption should
affect the
correlation
with
marketreturns
only
minimally,
if
at
all.
A
version
of
the
first-order
ondition which
permits
pjm
o vary
with
D,
while
retaining
the
assumption that
aD
0
,
is
presented as
equation
(4):
aP
aE(
)
acov(VI,
3D aD a a
3D
-
(V
a
OD
M
aCov(tJ
e
k)
When air " 0,
j # k,
equation (4)
reduces
to:
"A
number
of
well-known
real
world phenomena
are contrary to
this
assumption.
For
example, disability
income
claim filings are
negatively
correlated
with the business
cycle,
while
auto
collision claims
exhibit a
positive
correlation.
However, in
general,
a
reasonable
hypothesis
is
that pure
risk
occurrences
demonstrate a lower
degree of
interfirm
correlation
than
returns
from
productive
activities.
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
8/24
Risk Management
593
UP
aE(V)
S aa2tV()
CO?)
DD
DD
a0?m
D
j
)
Readers concerned about the assumption that =0 may wish to
9D
reformulate the
subsequent analysis
in
terms of
equation
(4)
or
(5).
An
importantobservation s
that in a
theoreticalsense the risk
manage-
ment
decision affects the
production
characteristicsof the firm;
i.e., it
alters the
probability distribution
of the
firm'stotal earnings at
Period
2.
Since
the
standard capital
asset pricing model
assumptionsare in
effect,
once the
risk
management
decision has been made,
how the
resulting
operations are
financed is
irrelevant; i.e., the mix
of debt and equity
financing has no impact on the marketvalue of the firm.9Thus, whether
the insurance
premium
and/or the retained losses
are paid for out of
equity or
debt-generated
financial resources is a
matter of
indifference
to the
firm's
shareholders.
Proportional
Retention
In
order to
explore further
the implications of
expression (2)
for risk
management
decisions, to define
E(V
)
and
Cov(VjVm)
in terms of
more familiar risk management variables seems appropriate.Thus, one
can
state that
v
=t
-
C
(6)
:1
'i 21
where
v'
W
net income of the
firm
considering
all
revenues and
all
expenses
other
than
those associated with
pure risk; and
C
=
the
pure risk costs of the
firm.
By
focusing on the
componentsof
Cj,
it is
then
possible to establish some
more precise decision rules.
In
this
section,
the
assumption
s
that
the firm can
base its retention on
a
quota
share
arrangement
on
"Original
erms."
n
that
case,
,-(1
.-c)G2
a(-
+
EJ)
(
where
G
-
the insurance
premium
which the
firm
would be charged for
complete,
first
dollar
coverage;
L - the
firm's
losses for
the
year;
Ej
=
the
cost of
administeringhe
pure
risk
of the
firm
f
all
losses
are
retained;10
nd
a
=
the
proportion
f the
risk
retained
by
the firm.
9
This
is
the familiar
Modigliani-Miller
Proposition
I.
For a
proof
of this
result
see
Fama
and
Miller
[5], pp.
160-164.
"'In
a
more general
formulation
Ej
would
have
fixed
and
variable
components with
the
latter
being a
function of
Li.
The
simpler
formulation is
utilized
here to
reduce the
computational
burden.
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
9/24
594
The Journal
of
Risk and
Imurance
Based
on
this
arrangement,
E&H
) - E(01)
-
l
-
a)G
+
aE(L;
)
+
]l
(8)
02(;t^-
U2(Bt)
+ 32cU2C)
- 2aCO"
.i;
(9)
:1
1
-
i'i
Note
that the insurance
premium
contributes
nothing
to the variance
of
the finnds
ncome
stream;
iLe,
by
purchasing
complete
insurance
coverage
the
firm can transfer
its entire
pure
risk
to the
insurer.
The
presence
of
the
covariance
term reflects
the fact that
losses
due
to
pure
risks
give
rise to indirect
as well as direct
costs. For
example,
a
business
interruption due to
a fire could result in
a decline
in
future
revenuesif some customers ail to returnwhen the firmresumesoperations.
The
subsequent
analysis
assumes
that all costs
of
purerisk
can be
subsumed
under
Lj
and,
therefore,
that
Cov(tLJ
0.
Based
on
that
assumption
and
equations
(2),
(6)
and
(7), the
first
order
condition for
a
maximum
in the
proportional
retention
case
is
the
following:
Oa2(j )
(G~j E(L;j)
-
EJ)
-
szpjm=_
Li
(10)
a
j
Solving
for
a,
one
obtains:
(G1a
Ei~
1
)s
~
(11)
a(j
)Smm
Verbally, tyis
directly
related to
the
marginal
reduction in
expected loss
costs
arising
from
an
additional
dollar
of
retention
(i.e.,
to
Gj-
E(tj)
-
En). (Note
that
retention
is not
feasible if this
term is
e
0.
Further-
more,
if
a
exceeds
1
the
analysis
suggests
that the
firm-
hould
become an
insurer.)
The value
of
a
also
bears a
direct
relationship o
the
ratio of
the
standard
deviation of
the
firm's
ncome
stream,
?(Vj),
and the
varianceof
its loss
cost
distribution.Thus,
the more
significant
the
pure risk
in
terms
of the
total risk
of
the
firm,
the
less of it
the
firm
should
retain.
Finally,
a
is inverselyrelated to the product of the marketprice of risk and the cor-
relation
coefficient;
.e.,
for
a
given
Sma
firm
whose
returns
are more
highly
correlated
with those
of the
market
should have a
lower
retention.
Aggregate
Deductibles
and
Self Insurance
Reserves
The
foregoing
model
should
be
relevant for
a
number
of
risk
manage-
ment
situations. For
example,
if
a
is
close to
1,
the
model
implies
that
complete
self
insurance
is
in
order.
Furthermore,
applying
the loss
unit
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
10/24
Risk
Managenwnt
595
concept and
making
the assumptionthat loss units
are independent,
the
decision
rule
could
be
applied separately
to the various pure risks facing
the firm.
For
instance,
the computation
may give
one answer for the
workers' compensation
risk and another for
that
of products
liability.
Nevertheless,many risk managementsituations exist to which (11) does
not apply.
If
a
is substantially less
than
1
but
greater than zero, for
example,
the results have
little practical meaning
as quota share
arrange-
ments
are
rarely
effected
between
non-insurance irms and insurers.The
more
common
arrangement,
of
course,
is the
use of
a
deductible;
and,
consequently,
the model
now will
be adapted
to
apply to the
deductible
selection
decision.
Aggregate
Deductible
Selection.The first case considered
s that
in which
the firmis faced with the choice of an optimal aggregate deductible;i.e.,
a deductible under
a
contract
in
which the insurer
agrees to pay all
losses
in
excess of a
total
annual
amount,
D. The retention
of the firm
under such
an arrangement
s:
L
-
,t
if
D;
and
Ji
i
*D
if
D
CL2)
AD
~ 1
,where
a5 the losses retained by
the
insured.
The
relevantparameters
can be
developed as follows:
H(t-)
J
dP(i)
+
DfdF(j)
(13)
D
N2R)
-
J
dFti
)
+
D2JdP(%)
E2(R)
(14)
ID
Differentiationwith
respect
to D
then yields (after
some simplification):
ID
w
-
F(D)
5
3D
2[
1
-
F
(D)]1[D
-
y~gj
Substituting these expressions
nto
equation (2), the first-order
condition
for
a
maximum,
one
obtains:
3G
[1
-
F(D)][D
-
Oi)+
Sp
}
(17)
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
11/24
596 The
Journal of
Risk and
Insurance
The left side of equation
(17) is derived by recalling equation (6) and
noting that
Cj
in this case
is
equal to the insurancepremium
plus
the losses
and
administrative
xpenses
retained under the aggregate deductible plan.
Although theoretically he optimaldeductiblecan be obtainedby solving
expression (17) for D, a
generalized explicit solution
to
the
equation
can-
not
easily be
derived.
Nevertheless,
this
version
of
the
optimization
condi-
tion suggests
two
important
observations about the use
of
aggregate
deductibles. First,
an
increase in the
aggregate
deductible
always
increases
the variance
of retained
losses; i.e.,
aa2(i)/laD
>
o.
This
relationship
can
i D
be verified,
under the
assumption
that
>
o
and
that
i .
>
o,
by
noting that
D
P
U(R
(t
dh)
+
D
UPt
D
<
H(t
)
+
DfdF(?)
- D.
0
D
Thus ,
D
-
E(?~R)
>
0
and, from
(16),
aa2CR)/aD
>
0.
Second,
it should be
evident
that a
necessary (although
not
sufficient)
condition
for
the
adoption
of an
aggregate
deductible
is
that
the left side
of
equation (17)
must
be
positive
for some
value
of
D. That
term,
of
course,
represents
the net
marginal
expected
savings
in loss costs
arising
from
the
use
of a
deductible; i.e.,
the
marginalpremium
reduction
aGJ/aD
less the
marginal
increase
in
the
expected retention,
1
-
F(D).
If the
expected increase in retained losses always is greater than the premium
reduction,
the firm has no incentive to adopt a deductible. Likewise, if
the term
is
zero; i.e.,
if
the
premium
is
actuarially
air
at the margin, the
firm
should not
adopt
the
deductible because
it
would be accepting a larger
variance with
no
offsetting
increase
in
expected return. The condition is
not
sufficient because a positive expected reduction in loss costs must be
large enough to offset the
resulting increase in the variance, with the re-
quisite relationshipbetween
the two
effects
determinedby the parameters
of the
capital asset pricing
model.
The Optimal Level of Reserves. Another interesting case involves the
choice
of an
optimal
buffer or
reserve
fund to
accompanya retention pro-
gram.
In order to
focus
exclusively
on
this
decision,
the
analysis is initially
conducted
under the
assumption
that
an
optimal aggregate deductible
already
has been
selected
and
that, given
that
deductible, the firm wishes
to
establish
an
optimal
size
reserve fund. The
case in
which the optimal
deductible
and
reserve fund
are established
simultaneously s considered
subsequently.
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
12/24
Risk Management
597
To
solve the
buffer
fund
problem
it is useful to redefine
Vj
as follows:
t e
Gj
(1
+
Rj) Bj
AdR
RC)
-
A
-Bj)
Rj
t
Bj
<
O
D
Two
additional variables have
been introduced:
Rj
=
the
rate of
return
on
shares of firm
j;
and
Bj
=
the
buffer or reserve fund for loss costs associated with pure
risks.
Equation (18) is equivalent to that utilized to derive the optimaldeductible
OV
#-%
(LJ-Bj)Rj, Bj
<
L
<
D
with the
exceptionof 2
terms,
Bj
(Rj-Rf
) and
{#
(D-Bj)Rj,
LiND
which are designed to represent the effect of reserves on the net
revenues
of the
firm. The
former term
appears
in the
equation
with a
negative
sign and represents
the
opportunity
cost
of holding reserve funds;
i.e.,
if
the firm maintainsreserves of
Bj*,
it
sacrificesearnings
of
RjBj*,
which
could be earned
if
those funds
were
invested
in its
production
processes,
but earns instead the lower amountRfBj*, which representsthe earnings
on
the reserve fund invested
in
relatively liquid,
risk free
assets.
The
latter
term, on the other
hand,
represents
the
firm's
sacrifice
in
earnings
if
losses
exceed
the
buffer
fund.
In
that
instance,
the
equation implies
that
the
firm loses its
earnings
for the
entire
year
on funds
paid
out
as losses.1'
That
aspect
of the
equation
constitutes a built-in
penalty
for
failure
to
maintain adequate reserves. That
is
the case because
the
equation
suggests that losses paid from reserve funds
are
paid at
the
end
of
the
year while
loss
payments
in
excess
of
reserve funds
are made
at
the
begin-
ning of the year. To understandthis point, note that for losses greater
than the reserves, lost interest
is
equal
to
Ri
times
the
excess
while
the
buffer
fund
is
assumed
to earn
interest
at
the
risk
free
rate for
the
full
year.
If
losses less than
the
buffer
fund
were
paid prior
to the
end of
the
year,
the
firm
could
not
be
credited with
the entire
amount,
BjRf.
A more
nearly
realistic version
of
the model
might
utilize
the
following
cost
ele-
ment
for
the buffer fund case:
-
-(Bj -
Lj) (Rf/
2) ,
L'j D
-
L
"The loss
paymentshemselvesrereflectedn L1.
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
13/24
598
The
Journal
of
Risk and
Insurance
This
expression
makes
the
alternative
assumption
hat losses are
paid, on
the average, midway
through
the year. Although
such an expression
is
more
realistic than the one
included
in equation (18),
the latter
formula-
tion has been utilized in the analysis. That decision was based on two
principal
considerations.
First, the desire to
compare
the results of the
present
model
with those obtained by
Duvall and
Allen
[4],
whose
formula-
tion is
consistentwith equation
(18) and second,
the difficulty
of fully re-
flecting the
consequencesto the
firm
of uninsured
ossesdifferent rom
those
expected,
due
to the abstract nature
of the
two period capital
market
model.
The
assumption
inherent in equation (18) helps
to remove this
limitation
as it
exacts a more significant
penalty
for losses
'in excess of
reserves
than the
alternative expression.1
For the reader
reluctant
to
accept this assumption, he optimalityconditionscan easily be derived by
performing
the
appropriatecomputations
under
the alternative
formula-
tion. These results
differmidetail but
not in substance
fromthose
presented
below.
Before
proceeding
with the computations,
it seems appropriate
to
re-
iterate
that the optimization
is carried out under
the assumption
that
Rj
and D are fixed; i.e., the optimal
buffer
fund will be derived
conditional
on
those
two variables.Taking expected
values
and
differentiating
equa-
tion
(18)
with
respect
to
Bj,
one
obtains
(after
some
simplification):
3B
0
)
- (j-
Rf)
+
RjEl
-
(Bj)J
(19)
This
result
is
equivalent
to that
obtained by
Duvall and
Allen
[4,
p. 501]
if one assumes
no taxes.
As
explained
in
their
article,
(Rj
-Rf)
represents
the
opportunity
cost to
the firm of increasingreserves by $1,
while
the
latter
term
is
the
expected
savings
of
the
firm
as
a result
of the
additional
dollar of reserves.
To obtain the
variance contribution
of the
buffer
fund, note that
when
Rj
and
D
are fixed,
the first three terms
of equation (18)
do not contribute
to the
variance
of Va. Differentiation
with respect
to
Bj
of the
variance
contribution
of
the last two
terms
yields 392B
Although
these
com-
putations
are
straightforward,
they are cumbersome;
and,
accordingly,
are presented in an Appendix,
which
is
available
from
the author. In
the
Appendix
it
also is demonstrated
hat
aac2(Vj)/aBj
<
0
for
0
< B,
D
that,
for a
fixed
deductible,
the firm can reduce its variance through
the
use of a
reserve fund.
An intuitive rationale
for
this
variance
reduction
is
that
earnings
be-
come
more
predictable
as a
result
of
the
earmarking
of
funds
for
the
12
The
penalty robably
s
realistic
ue
o
the
iquidity
roblems
hich
n
unexpected
lss
couldpresent.
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
14/24
Risk
Management 599
payment
of
unexpectedly large
retained losses. This function
of
the
reserve
fund
becomes
readily
apparent
when
one
thinks
of a
real-world
case in which
a firm can cushion the
impact
on
reported
earnings
of
an
uninsuredloss by drawing down its reserve fund. This increasedstability
in earnings might be expected
to
give
rise
to a
corresponding
tabilization
in the
price
of
the firm's
stock.
As
one might anticipate,
however,
the
use
of
a
reserve fund
generally
involves some
sacrifice of
expected
return
by
the firm. This
anticipation
is
the
case as
(19), although positive
for
small
Bj,
is
a
monotonically
de-
creasing
function of
that variable
and
can be
expected
to be
negative
for
a buffer fund of
any
substantial size. The
implications
of this result
for
the
selection
of
an
optimal
buffer fund
can
be
given
more
precise
form
by
setting up the maximizationcondition [from (2) and (19)]:
aa2(v
)fas
Rf
-
RF(B)
=
SmPjm
(20)
j 3 mjin
2a(Vj)
Since
au2(Vj)/aBj
<
0,
it is clear that if
[Rf
-
RjF(B,)]
is
positive fox
all
Bj
-?
D,
the buffer
fund
should be set
equal
to the
deductible.
In that
case the
use of
the
buffer fund would
both increase
expected
returnand
decrease the variance over the entire range of possible values of B,. In
view of the fact that
[Rt
-
RjF(Bj)]
may
become
negative for
B,
< D,
however,
it
seems
appropriate
o
examine
the most
likely
outcomes
under
that condition.
These
are
illustrated in
Figures la and lb.
Rf
E
(Vj)
]
Rf
E'(VY)
]
iB
optimum
8
ptimum
I
~~~~~~~~f
cr21t'1
~~~I
I
Figure
'a
Figure
lb
SELECTION
OF
AN OPTIMUM
RESERVE
FUND
FOR A PREDETERMINED DEDUCTIBLE
*V
In the
figures,
the curve
E'(Vj),
which
represents
[Rf
-
RjF(Bj)],
is
equal
to
Rf
when
Bj
=0, and
then gradually
declines,
crossing the B,
axis
for
B,
< D. The other
curve
f[Cr2'(V,)
,
which represents
aa2(0
)/aB
SMP
2a
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
15/24
600
The
Journal
of
Risk and Insurance
is below the
Bj
axis for all
Bj
>0 and
:
D and is drawn with a minimum
for some
Bj
< D.
As
demonstrated n the Appendix available from the
author,
the
curve is unambiguouslynegative, but the U-shape portrayed
by
the
figures is only
one
possible outcome. It is possible for the curve
to be monotonically decreasing; but if so, outcomes similar to those
illustrated n Figures la and lb still would representthe only alternatives.
In
Figure la, the value of the variance reduction resulting from
the
use of the buffer fund always is greater in absolute value than
E'(Vj).
In that case, the optimal solution would be at
Bj
=
D. Figure lb illustrates
the more likely outcome; i.e., the two curves intersect yielding an optimal
reserve fund less than the deductible. In both cases,
however,
the optimal
buffer fund occurs when
E'(MV)
< 0. This finding indicates that expected
value
decision
making
would result in a buffer fund of suboptimal size
as
in that case
the
fund is chosen where
E(VJ,)
=
.ys
Simultaneous Selection
of
Deductibles
and Reserves.
The
preceding
analysis has developed optimization
conditions
for an
aggregate
deductible
in the absence of
a
bufferfund
and for
a buffer
fund
given
a
predetermined
deductible.
While those decision
rules
were
useful
in
focusing
attention
on the individual
variables,
theoretical
precision requires
that a
more
general
result
be obtained
which
will
permit
the simultaneous
selection of
optimal
values for
the
decision
variables,
D
and
B,.
To
facilitate
the anal-
ysis, it is helpful to expressthe problemas follows:
Maximize:
Equation (1),
with
Vj
given by (18)
Subject
to:
Bj
D
Forming
the
Lagrangian,
Z
=
P,
-
k(Bj
-
D),
and
taking partial
deriva-
tives,
the
necessary
conditions for a maximumare obtained:
az i
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~Jaa2c
)fa
aZ
DE (V )
a
S
/aD
Q.
i~)
a
0
(21)
aD 1 + Rf 'D mjm 2a( )
9z
1
3E(R
)
aa2(~
)/aB
3Bj ' (B Smf) 2M
i
)
-
X
*
0
(22)
j f j
~~~~2a(~
X
>
0; B.
-
I)
<
0; A(B1
-
D)
-
0
(23)
Because
the constraint
s
an
inequality,
the
conditions
are
obtained
accord-
ing
to
the Kuhn-Tucker ules.
The
term
aE(V)
is
given by equation (19)
and
aa2(V)/9Bj
is
pre-
sented
in the
Appendix
available from
the
author.
However,
the
cor-
responding partial
derivatives with
respect
to
D
are different
from those
implied by equations (15)
and
(16)
because
those
expressions
did not
explicitly recognize
Rj
and
because
they
were
developed
under
the as-
2
Se
Duval
and
Allen
(41
p. 501.
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
16/24
Risk
Managenent
601
sumption that
no
reserve
fund would be used. With the
introduction of
those
factors,
it
can
be
shown
that:
DFE(v
G
an
a
_
a~:
(1 + R - - F(D)I(I + R) (24
and
-Z(9
-
2(1
+
Rj)t1
-
F(D)]{jL
dF(L)
+ jL(1
+
Ri)dF(i)
0 B;
+
BjRF (j
) - (1
+
R; )DF-(D) (25)
Under the assumptionthat
JdF(Ly
*
o and that the probability mass of
0
the
distributionover the
range
0
"I
<
D
is
not all concentratedat
Lj
=
D,
it
is
easy
to show that
aa2()/9D
>
0.
Furthernore, when BA=O, expres-
sion (25)
is
equivalent to (16) with the
explicit
recognition
of Ra. With
the introduction of a reserve fund,
aa2(t
)/aD
declines, and this decline
is
uninterrupted
or
O
8/9/2019 252028.pdf
17/24
602 The Journal of
Risk and
Insurance
Impact of Selected Assumptions
The preceding analysis of risk management decisions in the context
of
the capital asset pricing model has been conducted subject to a number
of simplifying assumptions. Many of these are no more restrictive in the
risk
management case than for other types of microeconomic analysis and
have not been analyzed further in
this
article on the ground that such
a discussion more properly belongs in an article on the capital asset pricing
model itself. However, two of the assumptions are particularly limiting in
the
risk management case, and the purpose of this section is to determine
the
effect of their relaxation on the risk management decision rules. The
two assumptions are: (1) that there are no taxes and (2) that the return
distributions are symmetric stable.
Taxes and Risk Management Decisions
The assumption about taxation is restrictive in the risk management
case because of the rules regarding taxation of casualty losses,
insurance
premiums, and contributions to self insurance reserves. Although
the tax
law in this area is
complex,
for theoretical purposes it is generally correct
to assume that
the
first
two items are deductible while
the
third
is not.
In
other words, the tax law contains
an
apparent bias
in
favor
of
insurance
and against self insurance.
The tax rules can be incorporated into the model by dividing Bj in
equation (18) by (1
-
t) where t is the corporate income tax
rate. This
operation
indicates
that,
when taxes
are present,
Bj/(1
-
t) dollars (before
taxes)
rather
than
Bj
dollars are needed to set
up
a buffer
fund of
size
Be. Carrying
out
the
optimization
with that modification
for
the most
general case discussed above yields a decision
rule
equivalent to equation
(26)
with terms
analogous
to
equations
(24), (25), 19, and
V2(Jv)
.
As
one
would expect, it can be shown that the introduction of taxes results both
in
a
smaller
buffer fund and a lower deductible than
in
the no-tax case;
i.e., the tax law does appear to be biased in favor of insurance.
Decision Rules with Non-SymmetricDistributions
The
significance
of the
symmetry assumption derives from the fact that
many
loss
distributions
are known to be positively skewed.14 However,
this
does not
necessarily invalidate the results obtained above utilizing
the capital market model. That is the case as the model is based on
distributions of total
return,
which have generally been shown to be
symmetrical.15 Apparently, the risk management aspects of the typical
firm's operations combine with its other activities in such a way that the
total
distribution retains the symmetry property. This observation,
in
fact,
is
what the central
limit theorem would lead one to expect for a
14
See, for
example, the
results
obtained by
Hartman
and
Siskin
[9].
15
Fama and
Miller
[51,
p. 260.
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
18/24
Risk Management
603
large
firm
with many (approximately)
independent activities. Another
factor which
may
limit the
impact
of the skewness
problem
is that
the
risk inherent in
the large loss tails of
loss distributionsusually is trans-
ferredby the firmto an insurancecompany.Thus, althoughthe distribution
of
retained losses
is
not
precisely
symmetrical,most of
the skewness effect
is
no longer present. As a consequence
of these factors, the foregoing
decision
rules
probably constitute a satisfactoryapproximationo the more
precise
theoretical results which would
be obtained if higher moments
of the loss
distributions
were
considered.
In
spite of
the foregoing considerations,optimizationrules are
developed
below which
recognize
the fact
that loss distributions
are
not
symmetrical.
These
rules are
presented for two majorreasons. First, it is possible that
for some firms the insurance capacity crisis, the general trend toward self
insurance,
and
the changing economic and legal environment will cause
risk
management decisions to play an
increasingly important role in the
determination
of
net
income.
Thus,
distributionsof returns for
those
firms
may
in
the
future acquire a more pronounceddegree of (negative)
skew-
ness. An
industry in which such a trend may be present is
the
drug
in-
dustry,
which
is
beset by burgeoning products liability problems.
The
second reason
for the development of a more general set of decision rules
is
to provide an indication of the nature of
the precise optimumconditions
in
the risk managementcase. This
precision is desired on the ground that
one
should fall
back
on
the central limit theorem
only as
a
last resort and
that even
then it would be helpful to
know the extent to which
mean-
variance
analysis
leads to a departure
from the exact result.
When the
symmetry assumption
is
removed,
the effect on
the
capital
asset
pricing
model
is
significant. If one
is
no
longer willing
to
assume
symmetry,one can no longer summarize
distributionsof
return
in terms
of their
means
and
variances
(assuming, of course, that utility
functions
depend on more than the first two moments of the return distributions).
Thus,
mean-variance
fficiency
ceases to
be
a useful
concept,
the
advantage
derived from the
assumption
of
risk
free
borrowing
and
lending
is
lost,
and
utility
functions
must
be
introduced
explicitly
into the
optimization
problem.
Unfortunately,
the introduction
of
utility
functions
gives
rise
to
other
fundamental
issues.
For
example,
not
immediately apparent is whose
utility function should
be introduced.
Consistency with the capital asset
pricing model
requires that the function
be that of an investor, but more
than one investor generally holds sharesin a firm and interpersonalutility
comparisonsusually are not acceptable.
Recent
developments
in
the
theory of finance
may offer a possible solu-
tion
to
this
dilemma.
Friend
and
Blume
[7]
have shown
that
the
relative
risk
aversion coefficient
of an individual investor can be
approximatedby
the following function:
16
16
See Friend and
Blume
[7], p.
10. Risk
aversion
coefficients are
discussed
in
Arrow
(2]
and
Pratt
[141.
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
19/24
604
The
Journal of
Risk and Insurance
11+1
Ck
where C - the harmonic mean of the
Ck,
and
ek
-
a disturbance term with zero mean.
In other
words,
a
reasonableassumption
s
that
in
terms of
expected
values
a
constant relative risk aversion coefficient for the
entire
securities
mar-
ket can be developed. Thus, to
utilize
a utility
function for the
entire
market in lieu
of one or
a series
of individual
investors'
utility
functions
may be acceptable.'7 This procedure
is
equivalent to assuming
that the
market, through
the
actions
of
a large
number of individual
investors,
arrives at a ranking of probability distributionsof return on risky assets
which possesses the properties required by the axioms
underlying
utility
theory.
The use of
a
marketutility function would retain
a
key advantage of
the
more restrictiveform of the capital asset pricing model; i.e., decision rules
could be developed and applied without constructing individual or cor-
porate utility functions.
As the use of the market ndex involves
a
number
of complex
problems,
however, that development has been consigned
to a future research project. This paper follows instead the more con-
ventional approach of adopting the utility function of the firm'smanagers
as
the
appropriate
index.
When
the
model
is couched
in terms of
utility analysis,
the
basic optimiza-
tion
rule becomes:
Maximize:
EMU(O)]
w
fU(j)dF(j) (28)
where
t-
he Period 2 value of the firm;
U(-)
-
the
utility function
of Period
2
value
at
the
beginning
of
Period 1;
and
F(')
-
the distribution
function
of Period 2 value of the
firm.
Expression (28)
indicates that the
objective
of
the firm is
to maximize
the
expected
Period
1
utility
of its Period 2
marketvalue. This objective
is
analogous
to the
capital
market model
objective of maximization of the
risk adjusted expected
Period
1
value
of the firm. One should recognize
that the discountingprocessis now implicit in the utility function although
that operation generally
will
be ignored
in the
examples presented below.
To
utilize
expression (28)
to
derive risk management decision
rules,
the
first
step
is
to
substitute
equation (6)
or
equation (18) for Va. Then
the
first order conditions
for a
maximum
are
obtained by setting the appro-
priate
derivatives
equal
to
zero. As the risk
management
maximization
s
"I
The
foregoing
onclusion
s that
of the
author
and
not
of
Friend
and
Blume.
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
20/24
Risk
Management 605
conditional
on
the other
stochastic
variables
which affect the firm and
because
those
variables
have
been assumed to be
independent
of
total
losses,
the expected value operation
can
be
carried
out with
respect
to
the
marginaldistributionof total losses ratherthan the complete distribution
of
Vs. For the
simplest case,
proportional
etention,
the general
rule would
be:
..L...
f
U'(V
)(G-
-L
-
E
)d(L
)
(29)
act
j j ij iJ
where
Vj
is defined
by equations (6)
and
(7). By
solving
that
equation
for
a
one would obtain the optimal level of proportional etention.
For
the aggregate
deductible-buffer
und
version
of the
model,
the
opti-
mization
problem
is:
Maximize: EE
U(J)
(30)
Subject
to:
BJ
<
D
By
~
~D
to
where
E[U(Y:)]
-
RJ(Yi)dH(?
+
fU(V2j)dF(j)
+
JU(V3J)dF(tJ);
o
B
I)
1
- G
(1
+ R;) -
BJ(RJ
-
Rf)
-
J;
V2:m
V1J
(i
- B
)Rj;
and
V3
-
,
+
Ad
-
D
-
(D
-
B
J)RJ.
The
maximization
would be carried out
according
to the Kuhn-Tucker
conditions with
the solution of
the
resulting equations
giving optimal
values for
Bj
and D.
In
order to provide an
illustration
of
the
nature
of the decision rules
generated by
the
utility
model,
the
maximization
problem is solved below
for
the
proportional retention
case under the
assumptions that losses
follow a gamma distributionand that the utility function of the firm is
exponential. The
use of an
exponentialutility
function
implies that the
firm's
absolute
risk
aversion coefficient
is
constant
and that its relative
risk
aversion
is
increasing.
These
properties mean that the
firm would
have the
same
tendency
to
insure
a
potential
source
of
loss of
a
given
size
regardless
of
its
wealth
and
that
if
the
magnitude
of the
insurable
loss
and
the
firm'swealth
increased
proportionally
t
would
be more
likely
to
insure
after
than
before
the increase.
While
one
may quarrel with the
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
21/24
606
The
Journal f
Risk
and
Insurance
reasonablenessof these
properties, it
is importantto recall
that risk man-
agement
decisions have only a
marginal mpact
on the wealth of the firm.
Thus,
for a
given firmat a particular
ime,
constant
absoluterisk aversion
probablydoes not constitute a serious departure rom realitywhile relative
risk aversion properties
are not
especially relevant. If one is
unwilling to
accept this
reasoning,
however, other
functions are
availablewhich possess
more
acceptable
properties.'8
With
gamma distributedlosses
and
exponentialutility,
the
proportional
retention
problem becomes:
Maximize:
E[U()]
y
(C
-
Cej)
r
(r?)r.eL
i
e
d
(31)
0
where
r--
the
absolute
risk
aversion
parameter,
and
A,
r
= parameters
of
the
gamma
distributionof
losses.
The
integrationeasily
is
accomplished
by recognizing
that
the
first
term
inside the
integral
integrates
to C while
the
second
can
be
written
as
-Ce--v
multiplied by
the
moment
generating
function of the
gamma
dis-
tribution, where V
=
V -
(l
-
)G1 -
cEj
and
the
parameter
of
the
mo-
ment generatingfunction is
ya.
Thus, one can write:
E[U(6'
)J
C -
Ce
Y(V)(l
-
Xa)-r (32)
j
A)
Differentiationwith
respect
to
a
yields:
BE1U(
)]
|-
.-Cey(v.
)
(-Y)
G
E)
(3:3)
-Bej(V)(.r(l
-
Y;frl(
The optimal solutionfor
a
then
becomes:
C1=
r
a
Y
(;
-
+
T
(34)
This
condition
indicates that the
level of retention
varies inversely
with the
degree
of
risk
aversion
displayed by
the
firm,
a result which is
both
reasonable and
consistent
with
intuitive
expectations.
Not so intui-
tively apparent is the reason that
a
varies inversely with (Gj
-
Ej). The
rationale for
this
relationship becomes clearer
when one
recalls that
retention
is feasible
only when
Gj
> E (
Lj)
+
Ej.
Thus, the
relationship
between
Gj
and
Ej
is
reflective of
the relativesignificanceof
the stochastic
and
non-stochastic
portions of the
total cost of the
retention program.
18
For
example,
the log
utility
function
exhibits
constant
relative
and
decreasing
absolute
risk
aversion.
See Pratt
[14].
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
22/24
Risk Management 607
Consequently, f
Ej
increases relativeto the total, one can
say in an intui-
tive sense that the riskiness of the
retention program declines
and that
a
can be set at a
higher level.
Empirical
Comsiderations
The
preceding
analysis has integrated risk management
variables
into
the theory of capital
market equilibrium. The resulting
decision rules
indicate that the
firm must consider more than expected
values when
developing its program
for dealing
with pure risk. As a theoretical con-
struct, the model
should be useful in focusing attention
on the
relation-
ships between expected
costs, risk,
and
other
parameters nvolved in
risk
retentionprograms. However, there appearsto be reasonto doubt its prac-
tical applicability.
One
problem
is the abstract nature of the model and
the number of
simplifying assumptionsemployed.
Thus, the decision
rules may be too
simplistic and may ignore too much
relevant informationfor real-world
application.A second
problem is that,
due to the magnitudeof the market
parameters and other variables appearing
in the model,
most of the
decision rules may
reduce for practical
purposes to expected value criteria.
An
interesting possibility
for future
research would be to test this conten-
tion through the use of real-world data. However, even if practical appli-
cation
is
not feasible,many of the foregoing
relationships till can be useful
for
organizing
and
analyzing risk management
data. Furthermore,he con-
struction
and use of utility functions appears to offer significant
potential
for
practical applications.
Summary and Conclusions
In
this article
the
capital asset pricing
model
of
the
theory
of
finance
has been adapted and applied to the development of risk management
decision rules. The
fundamental
concept underlying the analysis
is that
when
considering
a retention program,
a firm must recognize
not
only
the
savings
in
expected loss costs but also
the
increase
in
risk
accompany-
ing such a program.
The model reveals
that the firm should increase
its
retention to the point at which the marginal rate of substitution
between
expected
return
and risk is
equal
to the market price of risk multiplied
by
the
correlation coefficient between
the firm's
returns
and those of
the
market.
This
rule
was applied
more
specifically
for
the cases
of
propor-
tional retention and of aggregatedeductible selection. The latter problem
was solved both
with
and without the
use
of
a reserve
fund.
The
optimization
rules reveal that
the
degree
of
proportional
retention
is
inversely
related to
the ratio of
the variance
of
the
firm's oss
distribution
and
its
standard deviation
of total
return.
When
an
aggregate
deductible
is
adopted,
the
use of
a
reserve
fund
results
in a decreasein the
variance
attributable
to
the deductible.
The
recognition
of
this
negative
variance
contribution
mplies
that the
optimal
reserve fund should be chosen
where
This content downloaded from 200.16.5.202 on Wed, 11 Jun 2014 17:25:07 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
8/9/2019 252028.pdf
23/24
608
The
Journal
of Risk and
Insurance
<
0,
suggesting
that
expected
value
decision
making
leads
to
2,j
a reserve fund of suboptimalsize.
The decision
rules based
on the
capital asset
pricing
model can be
rendered
more
realistic by introducing
taxation and
by recognizing
that
loss distributions
are rarely
symmetrical.
When
taxes
are introduced,
it
can
be demonstrated
that the
optimal solution
implies
lower
values
for
both the aggregate
deductible
and
the
reserve fund.
The
relaxation
of
the
symmetry
assumption
necessitates
the adoption
of utility
functions.
Although the capital
asset pricing model decision
rules probably
have
limited
empirical
applicability,
utility analysis represents
a
potentially
fruitful alternativefor those who wish to avoid exclusive reliance on ex-
pected
values in
practical situations.
REFERENCES
1. Allen,
Tom C.
and
Duvall,
Richard
M. A Theoretical
and Practical
Approach
to Risk
Management.
New
York: American
Society of
Insurance
Manage-
ment,
1971.
2. Arrow, Kenneth
J. Essays
On the
Theory
of Risk Bearing.
Chicago:
Mark-
ham Publishing
Company,
1971.
3.
Borch, Karl. The Mathematical Theory of Insurance. Lexington, Mass.:
Lexington
Books,
D. C. Heath
and
Company, 1974.
4.
Duvall,
Richard
M. and Allen
Tom C. "Least
Cost Deductible
Decisions."
Journal
of Risk and
Insurance
40, no. 4 (December
1973):
497-507.
5.
Fama,
Eugene
F. and
Miller, Merton
H. The
Theory of
Finance.
Hinsdale,
Ill.: Dryden
Press, 1972.
6.
Feller,
William.
An
Introduction
to
Probability
Theory
and
Its
Applications.
Vol. 2.
2nd ed. New
York:John
Wiley & Sons,
1971.
7.
Friend,
Irwin
and Blume,
Marshall
E.
"The Demand for Risky
Assets."
Working Paper No. 1-74a. Philadelphia: Rodney L.
White Center
for
Financial Research,
Wharton
School,,
University
of Pennsylvania,
1974.
8.
Head, George
L.
"A
Capital
Budgeting Approach
to
Risk Management
Decisions."Paper presented
at the American
Risk
and Insurance
Association
A