Market Insurance, SelfInsurance, and SelfProtection Author(s): Isaac Ehrlich and Gary S. Becker Source: The Journal of Political Economy, Vol. 80, No. 4 (Jul.  Aug., 1972), pp. 623648 Published by: The University of Chicago Press Stable URL: http://www.jstor.org/stable/1829358 Accessed: 08/09/2010 21:54
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, noncommercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=ucpress. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a notforprofit 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 support@jstor.org.
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Political Economy.
http://www.jstor.org
Market Insurance, Selfinsurance, and SelfProtection
Isaac Ehrlich
University Chicago and TelAviv University of
GaryS. Becker
University Chicago of
The article develops a theoryof demand for insurancethat emphasizes the interaction between market insurance,"selfinsurance," The of and "selfprotection." effects changesin "prices," income,and othervariableson the demandfor thesealternative formsof insurance are analyzed using the "state preference" approach to behaviorunder uncertainty. Market insurance and selfinsurance shownto be subare stitutes,but market insurance and selfprotection can be complements.The analysis challengesthe notion that "moral hazard" is an inevitable consequence of market insurance,by showingthat under certainconditions lattermay lead to a reduction the probabilities the in of hazardousevents.
have usuallybeen to The incentive insureand its behavioralimplications analyzed by applyingthe expected utilityapproach withoutreference employedin consumption curve analysis ordinarily to the indifference expectedutility In is theory. thispaper insurance discussedby combining curveanalysiswithinthe contextof the "state prefand an indifference in (the preferences queserence"approachto behaviorunderuncertainty to tion relating statesof the world).' We use this framework restate to
Becker's contributionwas primarilyan unpublished paper that sets out the approach developed here. Ehrlich greatlyextended and applied that approach and was primarilyresponsiblefor writingthis paper. We have had many helpful comments from Harold Demsetz, Jacques Dreze, Jack Hirshleifer, and members of the Labor Workshopat Columbia Universityand the Industrial OrganizationWorkshop at the University Chicago. of 1 An approach originallydevised by Arrow (196364) and worked out in application to investment decisionsunderuncertainty Hirshleifer by (1970). 623
624
JOURNAL
OF POLITICAL
ECONOMY
and reinterpret a simpler moreintuitive in and way somefamiliar propositionsconcerning insurance behavior;moreimportant, derivea number we of apparently new results, especiallythose concerned with selfinsurance and selfprotection. Our approach separates objective opportunities from"taste" and otherenvironmental factors, whichfacilitates indean pendentinvestigation each class of factorsanalyticallyas well as of In empirically. addition,we considernot only the incentiveto insure, but also how muchinsurance purchased is undervarying "opportunities"2 and in viewof the existence the alternatives selfinsurance selfof of and protection. use the basic analyticaltools employedthroughout We traditionalconsumption production and theory. It has been argued that insuranceis different from"ordinary"goods and services because it is not desiredper se, but as a meansof satisfying morebasic needs.3Recent developments consumption in theory4 suggest, however, thatthe distinction between goodsand services purchased the in market and morebasic needstheysatisfy not a uniquecharacteristic is of insurance, but applies to all goods and services.The demand for the latteris also derivedfrom needs theysatisfy, the just as the demandfor factors production ordinary of in production theory derivedfrom is their contribution finalproducts. to The basic needsunderlying purchaseof insurance the will be identified withconsumption opportunities contingent upon the occurrence various of mutually exclusiveand jointlyexhaustive "states of the world."5 Market insurance this approachredistributes in incomeand, consequently, consumption opportunities, toward less wellendowed the states.Selfinsurance, however,redistributes income similarly,selfprotection a related has and effect, either might pursuedwhenmarket be insurance was not available. Moreover, optimaldecisionsabout market insurance dependon the availability theseotheractivities of and shouldbe viewedwithin conthe textof a morecomprehensive "insurance" decision.
2 Theorems concerning optimal insurancedecisions have been derived in two recent contributions Smith (1968) and Mossin (1968). Our approach differs by not only in formbut also in substance; for example, in the analysis of the interactionbetween marketinsurance, selfinsurance, and selfprotection. 3 For example, Arrow (1965) says, "Insurance is not a material good . . . its value to the buyer is clearly different kind from the satisfactionof consumer'sdesires in for medical treatmentor transportation. Indeed, unlike goods and services, transactions involving insurance are an exchange of money for money, not money for somethingwhich directlymeets needs" (p. 45). 4 See, for example,Becker and Michael (1970). 5 By consumption opportunities each state of the world is meant command over in commodities, producedby combining Ci, marketgoods, Xi, time spent in consumption, E ti, and the "state environment," , via householdproductionfunctions(for the latter conceptsee Becker and Michael 1970): Cij = fij(Xij, tij, Ei) j  1, . . . , m where i refers different to commodities. the productionfunctions If fullyincorporatethe effects of environment, utilityfunctionof commodities the would not depend on which state occurred. In particular,for an aggregate commodityC, U(C0) = U(C1) if C0 = 1' where 0, 1 denote different states.
INSURANCE
AND SELFPROTECTION
62
5
The first partof thispaper spellsout a modelof market insurance and discussesthe effects changesin termsof trade,"income,"and other of environmental factors on optimal insurance decisions. Selfinsurance, and selfprotection, a simultaneous determination the full insurance of decisionare thendiscussedin the second and moreoriginalpart.
I. Market Insurance
We assumeforsimplicity thatan individual facedwithonly two states is of theworld(0, 1) withprobabilities and 1  p, respectively, that p and his real incomeendowment each state is given with certainty IUe in by and 1,", where I,"  It,' is the prospective loss if state 0 occurs. If income in state1 can be exchanged income state0 at thefixed for in rate
d11
d1o
I'
(1)
state 1. The amountof insurancepurchasedin state 0 can be defined as the difference betweenthe actual and endowedincomes:6
s  Io  Io". (2)
it can be called the "price of insurance" measured termsof incomein in
The expenditure insurance on measured terms state l's incomeis in of b Ile
I I

_I, Sjt.
I("e),
(3 )
(4)
Substituting in (3) givestheopportunity (2) boundary
(I
or the line 21Bin figure It is assumedthat the individualchoosesthe 1.7 the optimalincomein states 1 and 0 by maximizing expectedutilityof the incomeprospect,
U*
(1
p) U(1i) +
p U(Io),8
(5)
The firstsubject to the constraint givenby the opportunity boundary. orderoptimality condition is p_ UO' (1
p)Ul'
(
(6)
6" Note that insuranceis definednot in termsof the liability"coverage" of potential losses, as in Smith's (1968) and Mossin's (1968) papers, but in terms of "coverage minuspremium," the net addition to income in state 0. or 7 In figure1, the opportunity boundaryAB is drawn as a straightline. This assumes that the same termsof trade apply to both insuranceand "gambling,"that is, to movements to the right and to the left of E, the endowment position. In practice, the opportunity boundary may be kinked about the endowment point. 8 For analytical simplicity we ignore the time and environment inputs and assume only a single aggregatecommodityin each state. Then the output of commoditiescan be identified with the input of goods and services,or with income.
626
JOURNAL
OF POLITICAL
ECONOMY
A
__
\
Certainty Line
E
/5AI
\pH
0
Ie
0~
~ ~~~~~~~B
FIG. 1
B
1
?
where(pU0') (1 p) U1' is the slope of the indifference curve (defined alongdU* 0), and n is theslopeof thebudgetline.In equilibrium, they mustbe thesame (see pointP). One can more completely factors separate tastes fromenvironmental by dividing (1  p) through (6) to obtain p in

1p p
_
U0' .
U
(7)
Further,i, the price of insurancedeflatedby the actuarially"fair"9 because a of price,pl(1  p), is a measure the "real" priceof insurance fairpriceis "costless"to theindividual(see the secondparagraph below). the Equation (7) thusimpliesthat,in equilibrium, real priceof insurance equals the ratioof the marginal utilityof I,) to that of I,, the ordinary demandtheory. resultin consumer curvebe conthat the indifference condition The secondorder requires at vex to the origin the equilibrium point,or D _p
declining."0
U0"f_
2 (1
_p)
Ulf/
> O.
(8)
A sufficient conditionis that the marginalutilityof incomeis strictly An immediate of implication equation (7) is that insurancewould be
9 An actuariallyfairexchangeis an exchangeof p/(1  p) units of income in state 1 for an additional unit of income in state 0, where p/(1  p) is the odds that state 0 would occur. 10Hirshleifer marginalutilityof (1970, p. 233) points out that althoughdiminishing income is not a necessaryconditionfor equilibrumat any given point,it is a necessary curve to be convex at all points. conditionfor the indifference
INSURANCE
AND SELFPROTECTION
62
7
demandedsome I, would be traded forIo if the slope of the indifference curveexceededthe priceof insurance theendowment at point,E:

U,'(10e)
U (I,.)
(9)
If the opposite were true, "gambling"would be demanded,provided similarterms tradeapply in redistributions incometowardstate 1. of of Note that gamblingcan occur withoutincreasingmarginalutility of incomeif the opportunities available are sufficiently favorable. Therefore, inferences about attitudestowardrisk cannotbe made independently of existing market opportunities: person a may appear to be a "riskavoider" underone combination pricesand potential of losses and a "risk taker" underanother." If the price of insurancewere actuariallyfair, equation (7) would reduceto 1  U(' U,': incomes wouldbe equalized in both statesof the world if the marginalutilityof incomewere always diminishing. This is "full insurance"in the sense thata personwould be indifferent to as whichstate occurred.'2 particular, small changesaroundthe equiIn for librium position, would act as if he were indifferent he towardrisk and interested onlyin maximizing expectedincome.Indeed,his incomein his 13 each state would equal his expectedincome; therefore, insurance fair can be regarded costlessto him.'4 as
11 Indeed, when faced with several independenthazards, a person might "gamble" and "insure" at the same time,provided the different hazards were associated with differentopportunities. For example, given a fair price of theftinsurance,he may fully insure his household against theftand at the same time engage in a riskyactivityif his expected earningsthere were greaterthan his earningsin alternative"safe" activities (see Ehrlich 1970). 12 Full insurance can be identified with full coverage of potential losses, since the equation I, I=) impliesthat I1e _b = 1se + d  b, where d is the grosscoverageand e Isle.By the same reasoning, b is the premium. Clearly,then,d = since an "unfair" price of insurance > 1 implies that I1 > Io, it also implies necessarily less than full coverage of potentiallosses. 13If I,0 I1=zI that is, I= Iies where Az=p/(1p), then =Io=Ioe+s, 14 ( 1p) Ile.
Althoughthe model has been developed for two states of the world, the analysis applies equally well to n states. We define the state with the highest incomesay, state nas the state without hazard and define all the states with hazard (h = 1, n  1) relativeto that state. Denoting by Ph the probabilityof state I, by
n1
I = pIO e+
p
1
Ph
h= 1
the probabilityof state n, and by Th the implicitterms of trade between income in state n and income in state h, it can easily be shown that if the termsof trade were fair (Jnh [P/P,] rsh= 1) Sh would be chosen to equalize incomes in all states of the = world and losses would be "fully covered." If the real termsof trade were unfairbut
628
JOURNAL
OF POLITICAL
ECONOMY
A. Substitution Effects The effect an exogenousincreasein the price of insuranceon the of demandfor Io, with the probability loss and the initial endowment of being the same, can be foundby partiallydifferentiating firstorder the optimality condition withrespectto t: 010
a
1 = Dj [(1
P) Ui' + (Io
Io")
r (1 P)
U1"].
(10)
the Since the denominator has alreadybeen shownto be positive, sign D or of equation (10) is the same as the signof the numerator, negativeif in Io > Io', since we are assumingU1" < 0. An increase the relativecost of incomein state 0 necessarily decreasesthe demandforincomein this state. Moreover, also reducesthe amountof insurance it purchased, since unchanged: Ioe remains aslan  aI(l)a  ajoean  ajola. Similarly, effect an increasein 3 on I,, and thuson the amount the of spenton insurance, is 011
,OS
1 D [1
_ p) Ulf + (Io oe) p Uoff
Joe) pUo"  spUO"
(1
Here theresult ambiguous is sinceUl' is positive whereas
(Io

is negativeif Uo" < 0 and s > 0. The resultis ambiguousbecause, although increasein n reducesthe amountof insurance an purchased, each unitpurchased becomesmoreexpensive. Consequently, amountspent the on insurance woulddeclineonlyif thepriceelasticity demandforinsurof ance exceeded unity15 proof obvious). (a is effect Equations (10) and (11) do not isolate a "pure" substitution because an increasein n lowersthe opportunities available (if s > 0). If bothI, and I( are superior effects goods, the incomeand substitution on bothreducethe demandforI,, whereas theyhave oppositeeffects the demandfor1I. Diagrammatically, the opportunity as boundarychanges P AB from to CD (see fig.2), the equilibrium pointshifts from to Q. If 1o werea superior good, Q mustbe to theleftof P'. Even if Io werean ininferior good, however, "pure" (that is, expenditurecompensated) a crease in the termsof trademustalways reduce the demandforIo and like S. increase theequilibrium mustshift from to a pointto its left, P II:
constant (7th = 1 + X> 1 for all h), sh would be chosen to equalize incomes in states with hazard only, that is, we would achieve what has been called full insurance above a deductible (for a definitionof this concept and an alternative proof see Arrow 1963). 15 This analysis,therefore, also shows that the effect a change in n on the "fullof ness" of insurance (the difference  Io) and thus on the degree of gross coverage I is generallynot unambiguous.
INSURANCE
AND SELFPROTECTION
629
C A E
S
P 0 D
FIG. 2
B
B.
Income Effects
Equation (4) can be written as
Ile + ?IoeW _I, ? + Io, (12)
whereW is a measureof the totalopportunities available. (This is shown in fig. 1 by the intercept on the I, axis.) The effect a changein OA of theendowments theincomedemanded each state can be determined on in by differentiating firstorder the condition:
aIo aW
aIo
aIo aIe
ajoe aie
aw
D3
D
(13)
aW
aIl
aIl adoi
ail" aw
aIl aIoe
D32
D
whereD;q _r(1  p) U1", and D)2 pUJ". The incomedemandedin each state necessarily if increaseswithopportunities the marginal utility of incomeis falling.Hence, an increasein each state's endowment increasesthe demandforincomein otherstatesas well. The effects the on demandforinsurance morecomplicated, are however, since theydepend on how different if endowments change.For example, Ile alone increased,
as
aIle

at1 i
azo0
(14)
630
JOURNAL
OF POLITICAL
ECONOMY
if and the demandforinsurance would increase.Similarly, Joe alone increased,
&9s
aloC
&O
&IO
1< 0
1
(15)
and the demandforinsurance would decrease.Equations (14) and (15) implythat if the difference endowedincomethe endowedloss from in a demandforinsurance would increase.Put differently,personwould be morelikelyto insurelarge ratherthan small losses (see Lees and Rice on of 1965).17 The effects a change in total opportunities the demand for insurancecannot be derivedwithoutknowledge the way opporof is tunities changeessentially because insurance a "residual" that bridges states the gap betweenendowedand desiredlevels of incomein different of the world.18 (and hence the size of the loss) are For example, both endowments if changed thesame percentage, by then
(88W
the hazard
increased either because Joe decreased or Ile increased, the
1)
(Io
0
1),
(16)
elasticities demandfors and I,, respectively.'9 of Equation (16) incorporates the ratherobvious conclusionthat the on depends effect a changein opportunities the demandforinsurance of on the effects the incomedemanded each state.If the slopes of the in on indifference curvesare constantalong a given ray fromthe origin (the indifference curvesare like EPF and GQ1H in fig.3)there is constant
16 According to equation (13), aI,/OI()e(1  p) U1,,]/[_pU("tt ;2 (1 [.2 p) Ui"] = 6, where clearly 0 < 6 < 1 if Ul" and U(" < 0. But since s  Jo Ioe,
where E,1w asl0W  Ws and
o Iol&W
* WlIo are the opportunity
17 Similarly,he would be less likely to take large gambles (see the discussion in Hirshleifer 1966). Of course,if insuranceis fair he will fullyinsure all losses, large or small. 18Note the analogy between insuranceand savings: the latter bridges the gap beat points in time. tween "endowed" and desiredlevels of consumption different 19 Given s _ and I1 C y 0I(e, then (Os/OW)(W/s) = (ds/dI(C) (I(C/s) = I(it)e 1 no (o/s)  (Ioe/s); by collectingterms,we get equation (16). Since IO s, eW if al > 1. L that is, if 10e = Ile If the loss is unaffected an equal increasein endowments, by where L is a constant,then se = (ds/dIle) (Ile/s) = (10/s)PT (d log W) /(d logIle) _(I1e/s). This impliesthat
as/lI()e
=

1 <0.
> e=
<
o
as
> ro=
< Io(I
W
+it)
> 1
(if Ii > lo)o
(16a)
INSURANCE
AND SELFPROTECTION
63I
C
II~~~~
o
B FIG. 3
D
0
positionslie on a givenray relative riskaversion20 thenall equilibrium  1. An equal 3, from origin, P and Qi do in figure and elf the as would thenincreasethe demand proportional increasein all endowments for insurance the same proportion. the slopes of the indifference If by
20
Note that
d slope
d do1
pUO p )u1'
y (UO"IUO');
d1o
subject to Ii = yIO is
] J
the latter definesin
0 as (U199/Ul')
constant, decreasingrelativerisk aversion.Similarly cr creasing,
d d1o
subject to I10
PU0'
L (1  p) U1' 
1
Lis L
0 as  (U199/U19) (UO"I/UO');
the latter definesin
creasing,constant,or decreasingabsolute risk aversion (see Pratt 1964, Arrow 1965). (Diagramatically,constant absolute risk aversion implies that the slopes of the indifference curves are constant along any 450 line joining two equilibriumpositionscurves are like EPF and IQJ in fig. 3.) Equation (16a) in n.19 the indifference implies that increasingrelative risk aversion, rO > 1, is compatible with decreasing absolute risk aversion,ese < 0, only if o < W/IO(l + Jc).
632
JOURNAL
OF POLITICAL
ECONOMY
and o and thus is relativeriskaversion, IQ2J in figure there increasing 3, relativeriskaversionimpliesthat the Fsw wouldexceedunity.Increasing
curvesincreasealong a givenray from origin, shownby EPF and the as
between and I, tendsto declineas opportunelasticity substitution of 1,) in elsewhere the ities increase.21 Regardlessof the shape of preferences relative(and absolute) risk aversionremains preference space, however, line. This constancyalways characterizes constantalong the certainty choiceswhenthe priceof insurance actuarially is fair (see fig.4). C. Rare Losses givenin equation conditions insurance for of An inspection the necessary of (9) shows that changesin p, the probability loss, do not affect the is incentive insureas longas the real priceof insurance independent to of p. If insurancewere actuariallyfair,the real price would always equal of unity, and thuswouldbe independent p. The deviationfroma fairprice,or the "loading" in insurancetermifrom identity the nology, be defined can
JtE
(1+FP'(17) )(17) 
111
~
~
~
Increasing relative
risk aversion
/'
Certainty Line: constant relative and absolute risk aversion
0
45
l0
FIG.
4
21 Since the slopes of the indifference curves necessarilyare constant along the "certaintyline" and by assumptionbecome increasingly steep toward I along other rays fromthe origin,a given percentagedeviation of the price of insurancefrom the fairprice resultsin smallerpercentagechangesin the ratio I1/I0 at higherindifference levels. That is, o= (dlogIl/Io)/dlog t decreases at higher indifference levels when nt equals the fair price.
INSURANCE
AND SELFPROTECTION
633
where is the "loadingfactor." X wereindependent p, so also would X If of be the real priceof insurance and p wouldhave no effect the incentive on to insure.In particular, therewould not thenbe a greaterincentive to insure"rare"lossesof a givensize.22 Since, apparently, rare losses are more frequently X insured,23 is presumablypositively relatedto p, perhapsbecause processing and investigatingcosts increaseas p increases.24 (An alternative is explanation provided by the interaction betweenmarketand selfinsurance analyzed in the nextsection.) Even if the incentive insurewereindependent p, to of theamount insured woulddeclineand the expenditure insurance on would increase p increased.25 as II. SelfInsurance and SelfProtection
Two alternatives marketinsurancethat have not been systematically to analyzed in the literature insuranceare selfinsuranceareduction on in the size of a lossand selfprotectiona reduction the probability in of a loss.26 For example,sprinkler systemsreduce the loss fromfires; burglaralarms reduce the probability illegal entry; cash balances of reducefluctuations consumption; in medicines, certainfoods,and medical
22 This result appears to contradict one by Lees and Rice (1965) because they definethe loading factorin termsof the gross rather than net amount paid in claim; that is by X' in Jt= [(1 + X') p]/[1  (1 + X') p]. A reductionin p, ' held constant, would reduce 2,our definitionof the loading factorand thus would increase the incentiveto insure. 23 Some evidence is presentedin Lees and Rice (1965). 24 Let the amount a that is spent processing and investigating each claim be the only administrative cost of providinginsurance.In a zero profitequilibriumposition, the unit price of insurancewould equal the ratio of the total amount collectedin premiums in state 1 (including administration costs) to the difference between the net amount paid in claims in state 0 and administration costs: A = (p d + p a)/[d (1  p) p a], where d is the amount covered by insurance.The degree of loading defined by X = = (d + a)/[dp ]1 = a/[dp (d + a)] would be [(1p)/p1 ;t1 a/(lp) larger the largerp was if d were fixed (d would tend to decrease as p increased,and this would increase ?i even further). 25 Generally, the effect an increasein p on the optimal values of If) and I , assumof ing that t = [(1 + X) p]/(i  p) and that i, le, and Ie are constant,is given by
0
1
ap
D
L
aill
ap
r
D lLoS
1"stip<u
1
t1
p i<
?
provided U" < 0. An increasein p would then lower the optimal amount of insurance s = IO  Ise and increase the optimal expenditureon insurance b  11e  I1. 26 These have been called "loss protection"and "loss prevention," respectively(see Mehr and Commack 1966, pp. 2829).
634
JOURNAL
OF POLITICAL
ECONOMY
checkupsreducevulnerability illness; and good lawyersreduce both to and the punishment crime. As these for the probability conviction of artificial distinguish to examplesindicate,it is somewhat behaviorthat reducesthe probability a loss from of behaviorthat reducesthe size of a loss, since many actionsdo both. Nevertheless, do so for expository we the insurance convenience and because selfinsurance clearlyillustrates principle redistributing of incometowardless favorablestates.
A. SelfInsurance
person as L
Assume that marketinsuranceis unavailable and write the loss to a  L(Lc, c), where Le  I1,  IoC is the endowed loss, c is L'(c) < 0. The expected
the expenditure selfinsurance, on and can utility be written as
U* ( 1 p) U(Ile C) +P
U(ICcL(LcC)
C).27
(18)
The value of c that maximizes equation (18), c0, satisfies firstorder the condition L'(cO) + 1 (1 p) U' *(19)
This maximizes expected utility the marginal if utility incomeand the of marginalproductivity selfinsurance decreasing, of are that is, if the indifference curvesare convexand if the production transformation curve between income states 1 and 0 (TN in fig.5) is concaveto theorigin.28 in A necessary condition a positiveamountof selfinsurance for obviously is L'(cO) > 1, or that therebe a net additionto incomein state 0. A sufficient if condition, the transformation indifference and curvesdo not have kinks,is that
1
L'(Le, 0) + p U' (Io) 29
1
(1

p) U' (fle)
(20)
An increase the unitcost of selfinsurance, in measured the marginal by
27 For analyticalconveniencewe assume that Ic alone is affected by c, although,of course,both endowmentsmay be affected. Moreover, the assumptionthat aL/ac < 0 is not always true: an individual could increaseIle and reduce If, by deliberately exposing himselfto hazards; for example,by committing crime or engagingin a risky a legal occupation (see Ehrlich 1970). The condition aL/ac > 0 can be said to define "negativeselfinsurance." 28 See equation (A5). Note that the transformation curve may be kinked at the endowmentpoint. 29 If the opposite were true, there would be an incentive to increase the loss by increasing and reducingI0 (see n. 27 above). I,
INSURANCE
AND SELFPROTECTION
635
C
A T Certainty Line E M
S~~~~
N~~~~~
450
0
FIG.
N 5
Io
wouldreducethedemandforselfinsurance, of productivity selfinsurance, by measured CO:30 ac0
<
o,31
(21)
wherea is a parameter that reducesthe absolutevalue of L' fora given a c. Similarly, reductionin Ifc would increase the demand for selfinsurance:
(22)
unlikethe to Equation (20) showsclearlythattheincentive selfinsure, to is incentive use marketinsurance, smallerforrare losses. The reason >0.32 is thattheloadingfactor selfinsurance largerforrarelosses because of is
30 Althoughc denotes the expenditureon selfinsurance rather than the reduction in the size of the loss, there is a onetoone relationshipbetween expenditureand insurancebecause L'(c) > 1. 31 By differentiating equation (19) with respectto cap, I(e and Ile held constant< one obtains ac(l/da = (pU('/U*( (() (OL'/Oa) (+)/() < 0, where U*(.(. 2U*IaC2 < 0 (see AppendixA), and by assumptionaL'/ac > 0. 32By differentiating equation (19) with respectto I(  ee, p and L' held constant, one obtains ac0/d0Ioc= acOl/Lc {pUo"[L'(c) + 1]/U*,,} aL/aLe >0, where by assumptionaL/OLe > 0.
636
JOURNAL
OF POLITICAL
ECONOMY
An increase endowed in in incomes thatresulted from investment human capital would probablybe associated with an increasein the marginal productivity selfinsurance.34 of on Therefore, effect selfinsurance the of a changein incomehas to be separatedfromthe effect the associated of changein marginal productivity. If marketand selfinsurance were both available, values of c and s wouldbe chosensimultaneously maximize expected to the utility function,
U*  (1

independenttheprobability loss.33 of of
its price,unlike the price of marketinsurance, to can be presumed be
p) U(Ile

c

s n) + p U(ll

L(Le,c)

c + s).
(23) If the priceof market insurance wereindependent the amountof selfof insurance, firstorder the optimality conditions wouldbe
(1
p)
U1'T+
pU0'_O.
(24)
+ 1] O.
(1
p) U1'  p Uo'[L'(c)
By combining theseequationswe get L'(c) +1
(25)
In equilibrium, therefore, "shadowprice"of selfinsurance the wouldequal the priceof market insurance. Clearly,marketinsuranceand selfinsurance "substitutes" the are in sensethat an increasein x, the probability loss beingthe same,would of decreasethe demandformarketinsuranceand increasethe demandfor selfinsurance.35 example, changein the market For a insurance line from AB to CD in figure5 would increaseselfinsurance the horizontal by distancebetween and M2 and reducemarket M1 insurance thehorizonby tal distance between and P. In particular, purchase market Q the of insurance wouldreducethe demandforselfinsurancecompare pointsS and, say, M1. Whenmarket insurance availableat a fairprice,theequilibrium is condition (25) becomes 1 P Ip ' I L'(c) + or (26)
33 The price of selfinsurance given by i =  1/[L'(c) + 1], where L'(c) preis sumably does not depend on p. The loading factoris then given by X = {1/[L'(c) + 1]} (1  p)/p]  1. Hence A/ap <0. 34 That is, not only would li~e/aE 0 i = 0, 1 whereE is the stock of human capi> tal, but probably also 02L/OcOE< 0. 35 A mathematical proof can be found in AppendixA.
INSURANCE
AND SELFPROTECTION
637
LI'(c) _,
preciselythe conditionthat maximizesexpected income.36 Even with diminishing marginal utility income, personwould act as if he were of a riskneutraland choose the amountof selfinsurance maximized that his expectedincome. Consequently, apparentattitudestowardrisk are dependenton marketopportunities, and real attitudescannot easily be inferred from behavior. More generally, evenif the priceof market insurance werenot fair,the optimalamountof selfinsurance would maximizethe marketvalue of income (given by W in equation [12]), and would not depend on the shape of the indifference curvesor even on the probability distribution of Geometrically, optimalselfinsurance determined moving states.37 is by alongthetransformation in figure to thepointof tangency curve 5 between thiscurveand a market insurance line; since the market value of income is the intercept the yaxis,thatintercept on wouldbe maximized such at pointof tangency. The effects specific of parameters the demandformarketand selfon insurancewhen both are available oftenare quite different fromtheir effect when marketinsuranceor selfinsurance alone is available. For example,although increasein the endowedloss increasesthe demand an forselfor marketinsurance wheneitheralone is available since an increasein market insurance itself reduces selfinsurance, vice versa,the and indirect effects can offset directeffects the whenboth marketand selfinsurance positive(see Appendix foran exampleof this). Similarly, are A becausea decreasein theprobability loss withno changein the market of loading factorreduces the demand for selfinsurance, increases the it demandformarketinsurance. Therefore, people may be more likely to use the market insurerare losses not necessarily to because of a positive relationbetweenthe probability loss and the loadingfactor(see the of in discussion SectionIC), but because of a substitution betweenmarket and selfinsurance. B. SubjectiveProbabilities, SelfProtection, and "Moral Hazard"
Selfinsurance and market insuranceboth redistribute income toward hazardousstates,whereas selfprotection reducesthe probabilities these of states.Unlikeinsurance, selfprotection not redistribute does income, because theamount spentreducing probability a loss decreases the of income
+j p [Ic

36Equation (26) can be derived by maximizing(1  p) (Ilc _ c)  L(LC, c)  c] with respectto c. 37 Equation (25) can be derived by maximizingW= (Ie  c) + IC  L(L', c) c] with respectto c. We are indebtedto Jacques Dreze for emphasizingthis point.
638
JOURNAL
OF POLITICAL
ECONOMY
in all statesequally,leavingunchanged absolutesize of the loss (its the relativesize actuallyincreases). Studiesusingthe statesoftheworld approachto analyzedecisionmaking under uncertainty have assumed that the probability a state is of entirely determined "nature" and is independent human actions. by of Withthisapproachthereis no such thing selfprotection; activities as the we call by thisnamewouldbe subsumed underselfinsurance. has been It claimedthat statescan always be defined guarantee independence to the of theirprobabilities from humanactions,38 we denythat thiscan be but done in a meaningful way. Consider, example,the probability for that a givenhousewill be damagedby lightning.39 Since thisprobability be can reducedby the installation lightning of rods,independent state probabilities could be obtainedonly by usinga morefundamental state description: the probability a strokeof lightning of itself.If controlof the is weather ruled out, the probability lightning be assumedto be of can unaffected humanactions.We are concerned, by however, about the probabilityof damage to the house we do not care about the probability of lightning se and the probability damage is affected lightning per of by rods. In other words,althoughan appropriatedefinition states would of produce state probabilities that are independent human actions, it of would not producea probability distribution outcomesthe relevant of probability distributionthat is independent theseactions.Since one of of the main purposesof the stateoftheworld approachis to equate the probability distribution outcomeswith the probability of distribution of states, a search for state probabilities of that are independent human actionswouldbe selfdefeating. To look at the difference betweenselfprotection selfinsurance and fromthe viewpoint outcomes, of assume the probability distribution of endowed outcomes givenby AB in figure Selfinsurance, contracting 6. by the distribution say, CD, lowersthe probability both highand low to, of outcomes,therebyunambiguously reducingthe dispersion outcomes. of on Selfprotection, the otherhand, by shifting whole distribution the to the left to, say, EF, reducesthe probability low outcomesand raises of the probability highones and does not have an unambiguous of effect on the dispersion.40 Since the preceding discussionshows that selfinsurance to be disis from tinguished selfprotection, developa formal we analysisof the latter. Let us assumethattheprobability a hazardousstatecan be reduced of by
:"'3This example is discussedby Hirshleifer (1970). 40)The effectof the introductionof selfprotection on the variance of income, I(C) 92, can be found by differentiation v'(r) =a Var(Ic)/ar  (1 (Ii p(Op)
38The only explicitdiscussionis by Hirshleifer (1970, p. 217).
INSURANCE
AND SELFPROTECTION
639
E
A
C
FIG.
D 6
F
B
ity of hazard,r is the expenditure selfprotection, plar on and p'(r) 0. If no market selfinsurance or wereavailable,the optimalexpendi<, tureon selfprotection would maximize
U*
appropriate expenditure: p 
p (p0, r), where pe is the endowed probabil
[1 p(pe, r)] U(Ie
r) + p(pe, r) U(Ioe
r);
(27)
theoptimality condition is

p'(rO) (U1

Uo)
(1

p) U1' +p Uo'.
(28)
The term theleftis themarginal on gain from reduction p; thaton in the the right, declinein utility the due to the declinein both incomes, the is marginal cost. In equilibrium, course,theymustbe equal. of The secondorder optimality condition requires that
U*rr

p"(rO) (U1

Uo) + 2p'(rO) (Ul'

Uo')
+ (1P)
U1"'+PU0"'<0.
(29)
Decreasingmarginal utility incomeis neither necessary of a nor a sufficient condition. p"(rO) > 0, equation (29) is always satisfied the If if marginal utility incomeis constant of and may or may not be satisfied if the marginal utilityis decreasing increasing. or This shows that the incentiveto selfprotect, unlikethe incentive insure, not so dependent to is
2p) (11_CI0e)2p'(0),
as p 1/2.
where r is the expenditure selfprotection. on Clearly v'(r) 0
640
JOURNAL
OF POLITICAL
ECONOMY
on attitudes towardrisk,and could be as strong riskpreferrers for for as riskavoiders. As withmarket and selfinsurance, effect a changein incomes the of on thedemandforselfprotection dependson thesourceof thechangeas well as on preferences.41declinein lIo alone might A not increasethe demand for selfprotection, if the marginalutilityof incomewere falling, even because a declinein Io" would increasethe marginal cost of selfprotecA declinein the marginal productivity selfprotectionan of increase in the shadowpriceof protectionalways decreasesthe demandforselfprotection regardless attitudes of towardrisk.43 Therefore, the endowed if probabilities incomeswerethe same, moreefficient and providers selfof protection would have lowerequilibrium probabilities hazard. Conseof quently, different persons different use probabilities theirdecisionmakin ingprocess onlybecauseof differences "temperament," optimism, not in or but also because of differences productivity selfprotection. sugin at As gested in the last section,differences productivity, turn,may be in in attributed differences education to in and otherforms "humancapital." of If marketinsuranceand selfprotection were jointly available, the function
U*
[1 tion.42
_p (pe, r) ] U (I,,,
r s 7t(r)) + p(pc, r) U (Iot r +s) (30)
would be maximized with respectto r and s; the firstorder optimality conditions are (1 p) Ul + p UO' (31)
 p'(r*)(U,Uo)
(1  p) Ul'[1 +s*z i'(r*)

p U'0 O.
(32)
The termit'(r*) measures effect a changein selfprotection the the of on priceof market insurance through effects p and the loadingfactor its on A.Fromthe definition n in equation (17) we obtain: of
r '(a) ='(p) p'(r) + ' (r). (33)
41 An equal proportional increasein endowments(Ile = yI0e) would increasethe demand for selfprotection (dr0/dIOC)= (1/U*,.) [p'(r ) (U1'y  U(') if conditionif U" < 0 is (U1'/U(') (II,,/Ie) + (1  p) U1" y + p U0s"]> 0. A sufficient ) 1, or that the "average relative risk aversion" between Ioe and 11C be sufficiently greaterthan one. 42
That is, ar(I/OId = [p'(rO) U 0

pU()"]/US
0 as[p'(rO) (?)/()
I/p
U. U/U ' 43 That is, rO0/la [(U1 sumption ap'/ap < 0.
UO)/U*,..] (Op'/af3)
> 0, where by as
INSURANCE
AND SELFPROTECTION
64 I
The effect marketinsurance the demandfor selfprotection of on has moralhazard refers generally been called "moralhazard." In particular, to an alleged deterrent effect marketinsuranceon selfprotection44 of thatincreases actual probabilities hazardousevents(Arrow1962). the of Consequently, moralhazard is said to be "a relevantcost of producing insurancethat is imposed by the insuredon the insurancecompany" of (Demsetz 1969,p. 7) and to providea "limitto the possibilities insurance" (Arrow1962, p. 612). We showedin the last sectionthat market insurancedoes reduce selfinsurance, no one has shown rigorously but insurance reducesselfprotection. why,or underwhat conditions, market on On Market insurancehas two oppositeeffects selfprotection. the one hand, selfprotection discouragedbecause its marginalgain is is betweenthe incomesand thus reducedby the reduction the difference of the utilities different in states (see equation [28]); on the otherhand, it is relatedto the is encouraged the priceof market if insurance negatively of amountspent on protection through the effect these expenditures on theprobabilities. of Consider relative the importance theseoppositeeffects in two extreme cases: If market insurance were always available at an actuariallyfairprice then p/(1 p), regardless the amountspenton selfprotection, at of and equation (31) impliesthat the optimalamountof marketinsurance (s*) equalizes incomein bothstatesof the world.There is still an incenbecause ;r is negatively related tive to spend on selfprotection, however, to theseexpenditures (r):
J~~t~r)
t (
( ) (1p)2 2.
(33a)
Substituting U1 U) and Ui' tion (33a) and the factthat5p'(r*)

U0 into equation (32), and usingequa(1  p) (Ic I( J), we get 1 (34)
'p _*
precisely condition maximize the to expectedincome.As withselfinsuran on ance, a fairpriceof market insurance encourages expenditure selfmoral hazard protection that maximizes expectedincome.Consequently, wouldnot thenincreasethe real cost of insurance, reducean economy's of insurance since an technical efficiency, limitthe development market or
44 See, for example, Arrow (1962, pp. 612, 613, 616; 1963, pp. 945, 961). Some writershave viewed moral hazard, in part, as a moral phenomenonrelated to fraud in the collection of benefits(see, for example, Mehr and Commack 1966, p. 174): a fireinsurancepolicy, for example, may create an incentivefor arson as well as for carelessness. Our analysis deals explicitlyonly with the effects marketinsuranceon of selfprotection, although implicitlyit applies also to the effects fraud. on
642
JOURNAL
OF POLITICAL
ECONOMY
amountof marketinsuranceequalizing income in all states would be chosen. Even moreimportant that,contrary the moralhazard argument, is to the optimal expenditure selfprotection, can be larger than the on r*, amountspentin the absenceof market insurance, By equations (34) r0. and (28), and thecondition p'(r) < 0, r* wouldbe larger thanr0 if
U (e
r0) U
Ile
oe
(Ie
r0)
U'
< (1
p) U'(Ile
r) + p U'(Ioe

r0),
(3)
whichis likelyprovided is not verysmall and U is concave.Indeed, if p r* utility werea quadraticfunction income, wouldbe largerthanrt if of p were largerthan onehalf.45 Not only are marketinsuranceand selfprotection complements the sense that the availability the former in of could increasethe demandfor the latter,but also in the sense that an increase theproductivity selfprotection a decreasein the real cost in of or of market insurance would increasethe demandforboth (see Appendix B). that the priceof marketinsurance Suppose,at the otherextreme, was independent expenditures selfprotection of on the loading factorincreased sufficiently offset to exactlythe reduction the probability in of loss. Selfprotection would thenusuallybe discouraged marketinsurby ance moralhazard would exist because the main effect introducing of market insurance wouldbe to narrowthe differences betweenincomesin different states.46 Moreover,since the demand for marketinsuranceis negatively relatedto the degreeof loading,it wouldbe negatively related to expenditures selfprotection. on for Consequently, thosekindsof market insurance withpricesthatare largely independent expenditures selfof on protection, shouldobserveeithera largedemandforinsurance one and a small demandforselfprotection, the converse. our judgment, or In this
45If U= aI+ b2, with b <0, equation (35) becomes [p (1/2)] Ie_ [p_ (1/2)] I(" > 0. Since Iie > I (6, thisimpliesthat p > 1/2. 46 If ;t'(r) = 0, the optimality conditionfor r, given the value of s, is fromequation (32): p'(r*) [U(I1 r*  sT)  U(IC  r* + s) ](1p) U' (I c r* s Js)  p U' (Ioe  r* + s) 0. Selfprotection would be discouragedby market insurance if an exogenousincreasein the latteralways reducedthe optimal value of r*; that is, if dr*/ds< 0, or dr*/ds= {p'(r*) [U1' (3T)  UO'] + [(1  p) U1" (Js) + pU0"1 }/ U*rr< 0, where U*rr< 0. The first termin the numerator necessarily is positive since p' (r*) is negativeand r, U0' and U1' are all positive. Therefore, sufficient a condition for the inequalityto hold is that the second term be nonnegative, since r = [(1 + or A) p]/(l  p), that U(" > (1 + X) U1"'. If X > 0no negativeloadingthis latterinequality necessarilyholds provided U" and U"' < 0; for example, if U were the quadratic function I + b 12, with b < 0. Of course,it mighthold even if U "' > 0. a
INSURANCE
AND SELFPROTECTION
643
is the majorreasonwhycertainkindsof hazards,like failure business, in are notconsidered insurable themarket. by Since the price of selfinsurance independent the probability is of of hazard (see the discussion Section11A) and thus of expenditures on in our insurance thatselfinsurance selfprotection, analysisof market implies is likelyto createa moralhazard.That is to say, the availability selfof insurancewould discourageselfprotection and vice versa. Moreover, in technological of progress theprovision one wouldtendto discourage the other. This analysisof moralhazard applies not only to the relationbetween selfprotection insuranceas ordinarily and conceived,but also to the relationbetweenprotection and insurancefor all uncertain eventsthat can be influenced humanactions.For example, unemployment by comdo pensation, relief, negative or incometax ratesincrease probability that the someone becomesunemployed? Does the presence underground of shelters increasethe probability that a country goes to war, the use of seat belts the probability an automobile of accident,or generous parentalsupport the probability that children become "irresponsible"? Since each of these,in effect, to of relatesa form insurance a form of protection, answersare not necessarily our "yes," and depend on how responsive the cost of insuranceis to the amountspent on protection. Sheltersand seat belts are ways to selfinsure, have costs that are and essentially unrelatedto the probability the hazards; therefore of they wouldtendto reduce(perhapsonlyslightly) incentive avoid a war to the or an automobile accident. theotherhand,if thecost in time, On embarrassment, etc. of applyingfor relief,unemployment compensation, or parentalsupportwere sufficiently positively relatedto its frequency, the answersmightwell be "no": the availability insurance of mightencourage theinsured makehis own efforts. to
Appendix A SelfInsurance and Market Insurance If both selfinsurance market and insurance available, the expectedutility are is
U*
(1 0
P) U(Iie
c  s ) + p U(Ile L(Le, c) c + s). (Al)
The values of c and s that maximizethis function must satisfythe firstorder optimality conditions
U*
U*C=

(1 0
(1

P) U11it pUot_ l+
p) U1'  p Uo'[L'(c*) + 1
8 
(A2)
0. (A3)
Clearly, equation (A3) would be satisfiedonly if [L'(c*) + 1] < 0: onlyif expenditures selfinsurance on increasedthe net incomein the hazardous state.
644 conditions are optimality Secondorder
U*ss(1 
JOURNAL
OF POLITICAL
ECONOMY
p) U1" JO + p Uo"1 < 0
(A4)
U*cc
(1
p) U1" + p UO"82 p (U*sc)2 >
Uo'L" <0
(A5) (A6)
=
L
U*S8 U*cc
O.
U" Equations (A4) and A5) are obviously satisfiedif everywhere < 0 and = 2LI0c2 > 0, that is, if the marginalutilityof income and the marginal are also These assumptions are of productivity selfinsurance both decreasing. to sufficient satisfyequation (A6) since U*Sc
A 
(1 
) ul"'
=
P Uo"'' < ?. 1, we can write
it2
(A7)
conditiont Utilizing firstorder the
 p(1  p) Uo' Ul" L
p2 Uo' Uo"L",
whichis positive if U" < 0 and L" > 0. A. Termsof Trade Effects of The effect an increasein it on theoptimalvalues of s and cIle, LC,and p held equations(A2) and (A3) withrespect constantcan be foundby differentiating to it. By Cramer'srule, ds*
AU=
dit

pU A
p( 1
1[
PAi+P
p) Uo'Ul'L" + p( 1

+
U P i
= < 0
p) Uo'Ul"s*jtL"]
(A8) + (1  p)U" s* i and  A2 (  P) Ul s* (1p)U1' where A1 are thepartialderivatives (A2) and (A3) withrespectto it. Similarly, of
dc* dit
A2U*ss  AU*sc
[ A
(1
~
A
p)2Ul'Ul"jt p)U 'U0"]
(
+p(1
>0. (A9)
can Hence, marketinsuranceand selfinsurance be consideredsubstitutes. of the By similarreasoning, effect an increasein p on the optimalvalues of is s and c, given that 2. in t = [(1 + 2.)p]l(1  p), Ile and Ile are constant, foundto be
ds* dp
BU*cc

B2U*c=
1
A
[UA
iLsj2+(
jt2+
p)U,'U1"
+ (1 p) Uo'Ul"  pUl'U0"1i + pU0'U0"82]also
< 0; (A10)
INSURANCE
AND SELFPROTECTION
645
(1
dc*
B2U*cc  BlU*8C
1 [
dp
A
(1

A
p) U'Uo1"t

pUl'Uo"
+
pUo'Uof8]
(
>
0
(A1)
where B1
and m'(P) = anlaP = T/Fp(l p)].
= (1  p)Ul"s*jt'(p)it,

B = U1'

U0' + (1 p)U1's*t'(p),
B. An Endowment Effect
The effect a decreasein I of be shownto be
Ile _
Ie, t, p, and L' held constantcan (+) (+)
(f)
ds* 9L p2U0'U0"IL" As* 02Uot~oo L L0t A aLe dIoe
where, assumption, by aLlaLe > 0; and
0
(A12)
dc* dld10' oe
SPUO8U*S pPUo" U*8c dL (A13) aLe =~~~~~~~~0, A A&
U*88. If the change in Ioe also
since by equations (A7) and (A8) U*s, =changed the results L', wouldbe different. Appendix B
SelfProtection and Market Insurance If both market is insurance and selfprotection available,the expectedutility are  [ 1 p(pe, r) U(I1e U* r s 7r(r)) + p(pe, r) U(Ioe r + s).]
(Bi)
The firstorder conditions are optimality
U*,   (1 O
U*r 
P) Ul

+ pUot_ 0(B2)
UO)

p'(r*) ( U
1  p) U1' [ 1 + s*jt'(r)
p pUo'
wherep'(r*) < 0 and m'(r*)< 0. Secondorder conditions that are
U*8 _ (1

0 O (B3)
p) U1"'2 + pUJ' < O

(B4)
+
U* rr
 P" (r*) ( U
UO) + (1  p) U1"[ 1 + s*r'(r*)2

pUo"
(B5)
+ 2p'(r*) { U1'[l + s*n'(r)
I U*ssU*rr

UO')
(1

p) Ul's*nt (r*) < 0. (B6)
(U*sr)2
>
0.
646
JOURNAL
OF POLITICAL
ECONOMY
Equations (B4) and (B5) would be satisfiedif U" < 0, if both p"(r*) and a"(r*) > 0, and if D = 2p'(r*){U1'[1 + s*jt'(r*)]  Uo'} (whichis positiveif U" < 0)47 were small in absolute value relativeto the othertermsin equation (B5). These conditions are also sufficient satisfyequation (B6) if, in parto D ticular, < (1  p)Ul's*jT"(r*) + p"(r*)(Ul  UO). Since it = [(1 + X)p]/(1 p), the effect an increasein r on it would be of
3t'(r)
(1 X) p'(r)
(1  p)2
+
pL'(r)
(1
(7
(B7)
where )'(r) gives the effecton the loading of an additional expenditure on selfprotection is generally and assumed to be positive.48 0 If insurance were always available at an actuarially fairprice,thenX(P) for all p; hence
it
p(r) 1p(r)
,
(r)
p'(r) (1_p)2 2 [p'(r) ]2
( p)3
n
and
i'(r)
Equation (B2) reducesto
(1
p"(r) p))2
>0 >
U1' = U0', and equation (B3) to
p'(r*)

(B8)
P)( and 1 + s* jt'(r*) it.
(B9)
Therefore, 3t'(r*) =1/[s*
(1p)]
A. Termsof Trade Effects
 p(r*)] If an initially fairpriceit =p(r*)]/Fl wereincreasedby an increase in the loadingwithno changein it'(r),49 the changein the optimalvalues of s and r would be givenby
ds*
dn(
ClU*rr
C2U*8r
Cl
)S L2_
2 Ul'1
 (1

<0,50
(BiG0)
47 According to equations (B2) and (B9) and the condition U" < 0, U1' [1 + sat' (r*)] < U0' if Jtp/1P). 48 One can write X'(r) = (ak/ap) (ap/ar, where ap/lr < 0. Hence, X'(r) > 0 only if ak/Op< 0. (But see our discussionin Section IC.) 49 Accordingto equation (B7), an increasein ?i due to an exogenousfactor 0 would not change t' (r) if, and only if, Ep'(r)/(1  p)] [Ok(r, )/80] =p [ak' (r, 0)/ 90]. This assumptionis made to separate an autonomous change in the price of insurance from an autonomous change in the effectof self protectionon the price of insurance. 50 Using equations (B8) and (B9) and the secondorder optimality conditions
INSURANCE
AND SELFPROTECTION
647
2(1 P)Ul'U*8r

dr*
cht
C2U*88
ClU*8r
( )
(+)
where
CU*S] 
, (B11)
(l

p)
U1'
+
(1
p) Ui"s*t
I8*,r*, p,
constant
7r'(r*)
and

C2
U*rw] r*,p, 7r'(r*) constant
8*,

p'(r*) U1's* + (
p)

Uj"s* [ 1 + s*rr'(r*)l

(1 p)U1'
(1
p)Ui"s*jt
(fromequation [B9]). Hence, if the price of insuranceincreasedfroman initiallyfairlevel,the demandforboth selfprotection market and insurance would decrease. If the price of insurance were always actuariallyfair,the effect an exogeof nous increasein the productivity selfprotection s* and r* withno change of on in the endowed probabilities and in the endowed incomes would be given by
ds* D2 U*8r
(+)
> ?
(B2)
and
dc* _
where D.)
D2 U*88
+
>0,
(B13)
 p)]1p'l0( and, by assumption ap' 0(3 < 0. TechUl'[s(l nologicalimprovements selfprotection thus seen to increasethe demand in are forbothmarket insurance and selfprotection.
B.
The Effect ExogenousChangesin p and L of
If insurancewere provided at an actuariallyfair price, and if the endowed probability increaseddue to an exogenousfactory withno changein p'(r), then
discussed above, it follows that U*rr= (1  p) U1"''2 + pUo0" + 2U1'/s* (1P) (1p) U1's*rT"' < 0; U*8r (r) U1"2pU0">0; and
I =
U*8
[_
LUlf
2
(
(1 1
p) Ul's* jt"(r)

>
0.
Since by equation (B4) U*ss < 0, 2Ul'/s*  (1 order for I to be positive.
p) U1' s* JT"(r) must be negativein
648
JOURNAL
OF POLITICAL
ECONOMY
ds* = dy
[(
(
P) U1i"s*TJn i(P)
]
Op
[U ()~~~
(1 p) Ul's*:n (r)
aY where, assumption, by > Op/dy 0, and
dr*
(+ )
'(p)[U*8s
+
( B 14)
U*sr] a
(1
P)Usl
S*t
0
(B15)
since U*_   U*',.5l The last resultis intuitively obvioussince a fairprice of insurance impliesthat  p'(r*)= r* of I,); therefore, is independent p providedthatp'(r) is unaffected changesin paY. the same reasoning one by By can showthatan increasein the size of theprospective loss increasesthe optimal values of both s and r. References Arrow,K. J. "Economic Welfare and the Allocationof Resources for Invention." In The Rate and Directionof InventiveActivity:Economic and Social Factors, edited by National Bureau Committee for Economic Research. Princeton, N.J.: Nat. Bur. Econ. Res., 1962. . "Uncertainty and the Welfare Economics of Medical Care." AER 53 (December 1963) :94173. . "The Role of Securitiesin the Optimal Allocationof Risk Bearing." Rev. Econ. Studies (April 1964) :9196. . Aspects of the Theory of Risk Bearing. Helsinki: Yrgb Jahnssonin Siitio, 1965. Becker, G. S. "Uncertainty and Insurance,a Few Notes." Unpublished paper, 1968. Becker, G. S., and Michael, R. T. "On the Theory of ConsumerDemand." Unpublished paper, March 1970. Demsetz, H. "Informationand Efficiency: AnotherViewpoint."J. Law and Econ. 12, no. 1 (April 1959): 122. Ehrlich,I. "Participationin IllegitimateActivities: An Economic Analysis." Ph.D. dissertation, Columbia Univ., 1970. Hirshleifer, "InvestmentDecision under Uncertainty:Applicationsof the J. State Preference Approach."Q.J.E. 80 (May 1966) :25277. . Investment, Interestand Capital. EnglewoodCliffs, N.J.: PrenticeHall, 1970. Lees, D. S., and Rice, R. G. "Uncertainty the WelfareEconomicsof Mediand cal Care: Comment." A.E.R. 55 (March 1965):14054. Mehr,R. I., and Commack,E. Principlesof Insurance.4th ed. Homewood,Ill.: Irwin, 1966. Mossin,J. "Aspectsof RationalInsurancePurchasing." J.P.E. 76 (July/August): 1968) :55368. Pratt,J. W. "Risk Aversionin the Small and in the Large." Econometrica32, nos. 12 (JanuaryApril 1964):12236. Smith, L. "OptimalInsuranceCoverage."J.P.E. 76 (January/February V. 1968): 6877.
51
See equation (B4) and the footnotefollowingequation (B10).
文档资料共享网 nexoncn.com
copyright ©right 20102020。
文档资料共享网内容来自网络，如有侵犯请联系客服。email:zhit325@126.com