Supply Response to Agricultural

Insurance: Risk Reduction and

Moral Hazard Effects
Bharat Ramaswami

This paper examines the consequences of agricultural insurance for expected supply.
The effect of insurance is shown to decompose into a "risk reduction" effect as well as
a "moral hazard" effect. The direction and magnitude of these effects depend on the
parameters of the insurance contract, producer's risk preferences, and the underlying
technology. Two models are considered for this purpose. In the first model, widely
employed in the literature, a producer controls only one input. The second model uses a
dual approach to extend the results to the case where a producer controls multiple
inputs.

Key words: agricultural insurance, moral hazard, risk, supply response.

A long standing issue in the analysis of agri- making.' In the words of Hazell, Pomareda, and
cultural insurance is supply response. Because Valdes, "Empiricalevaluation of the social costs
an insurance program alters the probability dis- and returns of publicly subsidized crop insur-
tributionof farm income, the question is whether ance requires measurement of the effect of risk
and in what manner producers adjust supply in reductionon supply responses. ... It is this risk
response to this change. The answer is of inter- response effect that leads to the major social gain
est to policy makers in developed and develop- from crop insurance" (p. 8). The study of sup-
ing countries, although for different reasons. In ply response is therefore important for an eval-
reviewing the history of crop insurance in the uation of agricultural insurance programs and
U.S., Kramerwrites "Research on price and in- consequently for the design of insurance itself.
come stabilization programs has indicated that In the shortrun, suppliesare alteredby changes
these programs have had supply response ef- in levels of variable inputs (input choice) and in
fects. As the crop insurance program becomes the allocations of fixed inputs (e.g., land) be-
a truly national program, similar effects may be- tween competing agriculturalactivities (activity
come evident from crop insurance, complicating choice). Crop insurance could affect both these
the supply control objective of commodity pro- choices. It is well known that insurance affects
grams" (p. 200). Developing country policy the incentives for input use. However, the di-
makers, on the other hand, place high priority rection of impact has been analyzed for a lim-
on expanding agriculturalsupplies. For them, a ited case only. Previous work has observed that
positive supply response is not a complication risk averse input decisions coincide with risk-
but a strong argumentfor publicly financed crop neutral decisions if insurance is complete (i.e.,
insurance programs which remove or minimize eliminates all risk), actuarially fair and contin-
the influence of risk on farm-level decision gent on input use. In this case, the effect of
insurance is straightforward. Compared to the
no-insurance world, the use of risk increasing
inputs increases while that of risk decreasing in-
BharatRamaswami is with the IndianStatisticalInstitute, New Delhi, puts falls, leading to an increase in expected
India.
An early version of this paper formed a part of my Ph.D dis-
sertation at the University of Minnesota. I thank Hans Andersson, 'For example, see Hazell, Bassoco, and Arcia for the objectives
Theodore Graham-Tomasi, Terry Roe, Arunava Sen, and an anon- of policy makers in Mexico and Panama. See also Lloyd and Maul-
ymous Journal referee for valuable comments and suggestions. This don for excerpts of the report of the Australian Industries Assis-
is University of Minnesota Experiment Station Article No. 20521. tance Commission, which discusses the objectives which should
Review coordinated by Richard Adams. guide government policy concerning agricultural instability.

Amer. J. Agr. Econ. 75 (November 1993): 914-925

Copyright 1993 American Agricultural Economics Association
Ramaswami Supply Response to Insurance 915

income (Ahsan, Ali, and Kurian; Nelson and that technology matters in determining the im-
Loehman). This case assumes that insurance pact of uncertainty (MacMinn and Holtmann,
payments are contingent on output realizations Pope and Kramer, Ramaswami). In particular,
as well as input choices, which for reasons ex- the property of technology that is relevant is
plained below, is a limiting assumption. How- whether an input is risk increasing or risk de-
ever, if the assumption is maintained, its im- creasing. The risk characterof an input has been
plication for activity choice is that complete useful in comparative statics analyses. For in-
specialization would occur in the higher value stance, researchershave used the restrictions on
riskier activity (e.g., Ahsan, Ali, and Kurian). technology implied by a risk increasing or risk
Such an outcome is confirmed in Hazell, Bas- decreasing input to predict the impact of public
soco, and Arcia who simulate the effect of crop policies on the use of environmentally hazard-
insuranceon croppingpatternsin a sectoralmodel ous inputs like pesticides and herbicides (Antle,
for Mexico. They find that an insurance pro- Leathers, and Quiggin; Olson and Eidman). The
gram that is actuarially fair and complete leads role of insurance has however, not been ex-
to higher expected production levels and a shift amined. Because the above studies documentrisk
towards riskier crops. to be an important consideration in the appli-
The desirability of complete insurance, how- cation of pesticides and herbicides, the impact
ever, depends (among other things) on whether of insurance may be significant. Consequently,
insuranceagencies are able to monitor farm level the results of the paper have a bearing on this
input use. If input use is not monitored, a pro- issue.
ducer who is completely insured will have little The second model considers supply response
incentive to apply any inputs at all. Then, it is when a producer can respond with changes in
not at all obvious that insurance would induce more than one input. The literature on produc-
greater specialization in the riskier crop. The tion decisions under production uncertainty is
situation where the insurer is unable to monitor generally confined to the one input case and has
input choices of insured farmers is one of moral little to say about the effects of uncertainty on
hazard. It is recognized as a major problem in the choice of input mix and the resultant impli-
the practice and design of agriculturalinsurance cations for average supply or its variance. The
(Chambers, Nelson and Loehman, Ray). The difficulty is two-fold. First, changes in input mix
widespread use of deductibles and the lack of depend on input substitutability and the distri-
complete insurance in real world situations is, bution of risk increasing and risk decreasing in-
in part, due to moral hazard considerations. A puts and calls for such knowledge of the pro-
realistic understandingof the effects of agricul- duction function.2Even then, comparativestatics
tural insurance must take moral hazard into ac- are harder because the use of all inputs is si-
count. This demands that empirical work on multaneously determined.
supply response be based on models of input use Second, the implications for expected supply
under moral hazard. and other parameters of the output distribution
A numericalmodel of input and activity choice is not straightforwardor immediate as this would
under moral hazard has been constructed by be in a single input model. For these reasons,
Kaylen, Loehman, and Preckel. The model is the second model employs a dual approach to
simulated for a hypothetical example with par- directly model a producer's choice of output
ticular specifications of utility functions, pro- distribution. In the usual primal problem, a pro-
duction functions as well as insurance pro- ducer chooses a vector of inputs to maximize
grams. While such models are necessary for expected utility of profits. However, if the first
delivering quantitative predictions, it is equally two moments completely characterize the prob-
importantthat the qualitativepredictionsof these ability distributionof output(Just and Pope), then
models be robust to small changes in numerical the producer'sproblem can be equivalentlyposed
specifications. In other words, are there any the- in terms of choosing a mean and a standardde-
oretical predictions of the effects of the crop in- viation. I show how the equivalence can be uti-
surance? In this paper I employ two models to lized to derive comparative statics about the
examine the supply response induced by insur- producer's choice of output distribution no mat-
ance through changes in variable input use. ter how many inputs form the input vector.
The first model considers the case where a
producer controls a single input. Such situations
of production uncertaintyhave been widely con-
2 I owe this
sidered in the literature. The principal insight is point to an anonymous Journal referee.
916 November 1993 Amer. J. Agr. Econ.

Model: General Assumptions contracts satisfying the above conditions is non-

empty.5
Farmersare assumed to be risk averse and max-
imize expected utility of profits where the utility
function U is increasing, concave and thrice dif- Insurance and Supply Response: A Single-
ferentiable. A stochastic technology is described Input Model
by a twice continuously differentiable produc-
tion function q(x, 0) where q is output, x is a Here a single input x enters the production func-
vector of inputs and 0 is a random production tion described by q(x, 0). An input is either risk
shock such that qO> 0. increasing or risk decreasing. If risk increasing,
An insurance contract consists of an indem- the marginal product 0) is monotonic, in-
qx(',
nity schedule I(q) and a premium P. Neither the creasing in 0 for all positive x. If risk decreas-
indemnity nor the premium is contingent on a ing, qx(', 0) is monotonic, decreasing in 0 for
particular choice of input vector. Given a con- all positive x. The terminology derives from the
tract {I(q), P} a producer chooses input appli- fact that higher levels of input use lead to higher
cation x(I, P). When no insurance is purchased or lower output risk depending on whether
qx0
or is available, x(O, 0) denotes the producer's is positive or negative. As noted later, the op-
optimal input level. The initial situation is as- timal application of a risk increasing (decreas-
sumed to be one of no insurance. This is com- ing) input under risk averse preferences is al-
pared with a final situation where an insurance ways less (more) than the risk-neutral level of
contract {I, P} is available. The question is, how input use. In an agricultural context, fertilizers
does the purchase of insurance alter expected are often risk increasing in their impact on out-
output? The answer involves a comparison of put risk (e.g., Just and Pope) while pesticides
Eq(x(I, P)) and Eq(x(O, 0)). and herbicides are risk decreasing (e.g., Antle,
The comparison is carried out for the set of Olson and Eidman). A more general interpre-
insurance contracts satisfying the following con- tation, pursued in Lewis and Nickerson, is to
ditions:3 regardx as expenditures on self-insurance. Then
(i) Feasibility: Given the optimal input re- such expenditures are risky if qo > 0, but are
sponses of farmers, insurance is actuarially fair, risk-reducing if qx0 < 0. Many examples of such
i.e., P = E[I{q[x(I, P), O]}]. Contracts satis- expenditures are provided in Lewis and Nick-
fying this condition are feasible. The premium erson.
on a feasible contract is equal to the expected Let ir(x, I, P) be the producer income as a
level of indemnity. function of input level and the insurance con-
(ii) Differentiability: The indemnity schedule tract. Then IT(x, I, P) = q(x, 0) - wx + I(q)
is differentiable everywhere except possibly at - P, where w is the input price and 7r, q, w, I
a finite number of points. This condition is usu- and P are all normalized with respect to a cer-
ally met by real world contracts.4 tain output price. The analysis easily extends to
(iii) Monotonicity: The indemnity schedule is the stochastic price case by regarding q as rev-
monotonic decreasing in output (or monotonic enue ratherthan output. Naturally, the interpre-
increasing in loss), i.e., I'(q) O0.Strict mono- tation of the model is affected,6 but not the re-
tonicity is not required. In other words, insur- sults.7
ance paymentsare larger(or at least, not smaller)
for larger losses. This assumption is also con-
sistent with real-world insurance practice.
Under mild regularity conditions, the set of 5 Pick a bounded indemnity schedule which is differentiable
and monotone decreasing. The feasibleI(') premium consistent with
I(q) is the fixed point of the equation: P = EI(q(x(I, P), 0)). Let
3 Alternatively, the form of the insurance contract could have g(y) = EI(q(x(I, y), 0)). g is continuous in y. Let I, and 12 be the
been endogenously derived (Mirrlees, Holmstrom, Chambers 1989). lower and upper bounds of the indemnity schedule. If y E [I, 1,],
However, as noted by Hart and Holmstrom, optimal contracts im- g(y) E [I,, 12]. Then by Brouwer's fixed point theorem, there exists
a P E [1,, 12] satisfying P = g(P).
pose little structure on the form of the insurance contract. For in- 6
stance, the literature is not able to deduce that optimal insurance First, if R(x, 0) is the gross revenue as a function of input level
contracts are increasing in the magnitude of loss. and a random shock, input x is risk-increasing (or risk-decreasing)
4 Consider a U.S. Federal crop insurance contract where if R, is increasing (or decreasing) in 0 for all x. Second, the in-
surance program insures revenue and not output.
-
I(q) = r(q* q) if q q*, r > O0 7As this paper considers comparative statics with respect to ar-
= 0 otherwise. bitraryinsurance contracts, the model is unaffected by results which
show the sensitivity of optimal insurance schemes to output price
I is differentiable everywhere except at q* risk (Ramaswami and Roe).
Ramaswami Supply Response to Insurance 917

Complete Insurance s(x) = (Eqx(x,0) - w)/Eqx(x, 0)

Given an insurance contract, the optimal choice =

oMu(x, I, P) var(U'(IT(x,I, P))
of x is dictated by
/EU'(OT(x,I, P))
(1) = and
maxx r(x, I, P) = EU[q(x, 0) - wx + v(x, 0)] Mp(x) /var(qx(x, 0))/Eqx(x, 0)
p(x, I, P)
subject to x ? >0 I, P)), qx(x, 0))
where v is the payoff from insurance, i.e., v(x, -cov(U'(ir(x,
0) = I[q(x, 0)] - P and r is the expected utility
of the producer. With insurance, the change in \ivar(qx(x, 0)) Vvar(U'(T(x, I, P))
expected utility as a result of a marginal change s is the fraction of expected marginal product
in input use is that the producer receives as income after sub-
tracting input costs. oMuand OM are the coef-
(2) rqx(x,I, P) = EU'[iT(x,
- {[1 +
I, P)] ficients of variation of marginal utility and mar-
I'(q)]qx(x,
0) - w}. ginal productrespectively and p is the (negative)
An insurance contract is complete if I'(q) = correlation between marginal utility and mar-
- 1 for all q, i.e., if insurance fully compensates ginal product. Using these definitions,
an incremental loss. For such a contract,
(5) qx(x, 0, 0)/ EU'(ir(x, O))Eqx(x,0)
= s(x) - p(x, 0,
71x(X,I, P) = -wEU'(7T(x, v)) < 0 O)cMU(x,0, O)oM(x)
for all x ? 0. Thus complete insurance provides (5) expresses the effect of input use on ex-
no incentives for positive levels of input use. pected utility as the difference between a mean
The rest of the paper assumes that the optimal effect (the first term) and a risk effect (the sec-
level of input use is positive. This means in- ond term). By the first order conditions that
surance is incomplete, i.e., I'(q) -1, where characterizethe optimum, the two effects are just
- q.
the weak inequality is strict at some equal. At the risk-neutrallevel of input use, the
expected marginalproductis equal to input cost.
This means s(x) is zero and hence so is the risk
No-Insurance Case effect. Thus, whether risk pverse farmers use
more or less input than the risk-neutrallevel de-
In the initial situation, I(q) = P = 0 for all q, pends on whether the risk effect is negative or
and the incremental change in expected utility positive.
due to input use is The sign of the risk effect is determined by
p, which is the negative of the correlation be-
(3) tween marginalutility and marginalproduct. For
71x(x,0, 0) = EU'(QT(x,0, 0))(qx(x, 0) - w). concave utility functions, marginal utility is
monotonic decreasing in 0. On the other hand,
Because cov(U', = - can
qx) EU'qx EU'Eqx, (3) the marginal product is either monotonic in-
be rewritten as
creasing or decreasing in 0. It follows that the
(4) qx(x, 0, 0) = EU'(7T(x,0, 0))(Eqx(x, 0) risk effect is positive if the input is risk increas-
- w) + cov(U'(Tr(x, 0, 0)), qx(x, 0)). ing and negative if the input is risk decreasing.
Consequently, risk averse level of input use is
Dividing by EU'(rT(x, 0, O)Eqx(x, 0), qxcan be greater (smaller) than the risk-neutral level of
expressed, with its sign preserved as a number input use if the input is risk decreasing (increas-
independent of the units in which output and ing). This result has been noted earlier in
utility are measured. MacMinn and Holtmann and in Pope and Kra-
mer.
0)
77g(X, = (Eqx(x, 0)O,- O))Eqx(x,
O, O)/EU'(wT(x,
w)/Eqx(x, 0)
+ cov(U'(Ir(x, 0, 0)),
Risk Reduction and Moral Hazard Effects
qx(x, 0))/[EU'(wT(x,O, O))Eqx(x, 0)]
The above can be expressed compactly with the A decomposition similar to (5) for rx (x, I, P)
following notation. Let reveals
918 November 1993 Amer. J. Agr. Econ.

(6) I, P)/EU'(Tr(x, I, P))Eqx(x, 0) O)OM(xi) if the input is risk-increasing. (ii) 0 >

=qx(x,
s(x) - [p(x, I, P)o-MU(x, I, P)o-MP(x)] I, P)p(xi, I, P)rMp(xi) >
O'Mu(xi, OMu(Xi, O, O)p(xi,
- [-EU'(IT(x, I, 0) 0, if the input is risk-decreasing.
O)uMp(Xi)
P))vx(x,
I,
/[EU'(iT(x, P))Eqx(x, 0)]]. Proposition 1 is proved in an appendix. As
Compared to (5), there is an additional moral intuitionmight suggest, proposition 1 asserts that
hazard effect represented by the third term in actuarially fair insurance reduces the absolute
(6). When a producer buys insurance, a change value of the risk effect of input use. By itself,
in input levels, besides affecting output, also af- this leads a producer to adjust input application
fects indemnities. Because an increase in output in the direction of risk-neutrallevels, i.e., to in-
reduces indemnities, the marginal return to an crease input use if risk increasing and to de-
additional unit of input application is the mar- crease input use if risk decreasing.
ginal product of that additional unit less the re- However, the sign of rl(x0, 0, 0) also depends
sulting loss in indemnities;i.e., it is qx(1 + I'(q)) on the sign of the moral hazard effect which is
that is less than qx. The resulting change in ex- always positive since Pv = I'(q)qx < 0 for all x
pected utility from the loss in indemnities is and 0. As remarked earlier, insurance reduces
EU'(Or(x,I, P))qxl'(q)), which is the numerator the marginal return from an additional unit of
of the third term in (6). input application which, therefore, reduces in-
To simplify notation, let xi = x(I, P) and xo put use irrespective of whether it is risk reduc-
= x(O, 0). To compare xo with xi, assume that ing or risk increasing.
,q(x, 0, 0) is concave in x.8 Then xi Mxo as rx(xi, Because the moral hazard and the risk reduc-
0, 0) 0 and so checking the latter condition tion effects are in the same direction for a risk
-
is sufficient. Because EU'(Tr(xi,0, 0)) and Eqx(x,, decreasing input but of opposite directions for a
0) are both positive, rx(xi, 0, 0) is of the same risk increasing input, crop insurance has differ-
sign as rx(xi , O)/EU'(T(xi 0,O, ))Eqx(x, 0). ent impacts in the two cases. When an input is
From (4), risk decreasing, both the risk reductionand moral
hazard effects are positive which makes qj(xi,
(7) rx(xi, 0, 0)/ EU'(-r(x;, 0, O))Eqx(xi,0) 0, 0) > 0. When an input is risk increasing, the
= -
S(Xi) OUM(Xi, 0, O)OMP(Xi)P(Xi,0, 0) risk reduction effect is negative but the moral
hazard effect is positive, which makes the sign
But from the first order condition, xi solves
of x(xi, 0, 0) indeterminate. The next result is
= I, P)
I, P)OrMp(X)p(xi, therefore immediate.
s(xi) oMau(X,
+ [- EU'(iT(xi, I, P))vx(xi, 0)/ 2: With all constant and de-
PROPOSITION
[EU'(T(xj, I, P))Eqx(xi, 0)]]. creasirig risk averse utilityfunctions, the impact
Substituting for s(xi) in (7), of actuarially fair crop insurance on input use
is (i) to reduce it if the input is risk decreasing
(8) rx(xg,O, O, O))Eqx(x,,0) and (ii) indeterminate if the input is risk in-
= [OuM(x,
O)/EU'(-r(x,,
I, I, P) creasing.
- P)p(xi,
0, O)p(X , 0, 0)]
UrMU(X0, The implications for supply response are
MP(Xi) + [- EU'(IT(xi, I, P))vx(xi, 0)/
[EU'(wT(x,,I, P))Eqx(xg,0)]]. straightforward.If the input is risk decreasing,
insurance affects its use in a manner that de-
The first expression in square brackets above is creases expected output and increases the risk-
the difference between the risk effect with in- iness of output. However, if it is risk increasing,
surance and the risk effect without insurance,
(expected) agricultural supply and its riskiness
both evaluated at x,. Because of the following may increase or decrease depending on the rel-
result, the differencebetween risk effects is called ative strengths of the moral hazard and risk re-
the risk reduction effect. duction effects.
PROPOSITION 1: For all constant and decreas-
ing risk averse utility functions and for a fea-
sible insurance contract {I, P}, (i) O < aMu(xi, A Numerical Simulation
I, I, P)ap(xi) < O, O)p(xi, O,
P)p(xi, Mu(xi, Because the theoretical analysis is inconclusive
in the case of a risk increasing input, a simu-
8
Because thx(x,0, 0) = EU'(Qn(x,0, O)(qx(x,0) - w)2 + EU'(ir(x,
0, O))q,(x, 0), concavity of the production function together with lation experiment is undertaken to obtain in-
risk aversion is sufficient for concavity of expected utility in x. sights. Consider a specialization to linear insur-
Ramaswami Supply Response to Insurance 919

ance schedules, constant risk averse utility Note that if the mean elasticity is much larger
functions, and normally distributed output risk, than the product of variance elasticity and rel-
i.e., U(y) = - exp(-Ay), q(x, 0) = p(x) + ative risk premium, r* could be negative. This
o(x)0 and I'(q) = -r, where A is the constant means insurance would always decrease input
coefficient of absolute risk aversion, 0 is nor- use. For instance if p = 0.5, r* is negative
mally distributed with zero mean and unit vari- whenever the mean elasticity is greaterthan twice
ance, and 0 <: r < 1. x is risk increasing, i.e., the variance elasticity.
qxo = x> 0. Figures 1 and 2 plot the outcome of a nu-
The optimal input choice is found by solving9 merical experiment. It is assumed that pi(x) =
xa and or(x)= xY. The optimal value of x is plot-
(9) ted against 20 values of r between 0 and 1. The
(1 - r)Ax(xi) - w - (1 - = 0. calculations use equation (9) and assumed val-
r)2Au(xi)ox(xi) ues of w = 0.05, and A = 0.2. In figure 1, the
Note that xi is independent of P Hence we can
write xi = x(r) to denote the functional depen- elasticities, a and 3, are the ones estimated by
dence on r. Similarly xo = x(O). Just and Pope for a data set on the output re-
The decomposition into risk reduction and sponse of corn to nitrogen application. Here the
moral hazard effects can be derived as mean elasticity is nearly three times the variance
elasticity. Consequently, insurance decreases
(10)
0, O)/EU'(-r(x(r), O, O))Eqx(x(r), 0)
rhx(x(r),= (r2 - + r Co
2r)Ao(x(r))ox(x(r))/1,(x(r))
from which, we derive

(11) x(r) - x(O) as r r*

where r* = 2 - Ax(x(r))/Ar(x(r))a,(x(r))
I-3

that is, insurance increases input use if insur- a O

ance is "limited"; otherwise input use declines
relative to the zero insurance level. r* is the crit-
ical value of insurance such that, if r is less than
r*, input use increases but if r is greater than
r*, input use declines. The result is consistent
with intuition as one would expect the moral 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
hazard effect to be relatively stronger at larger Insurance
levels of insurance.
The critical value r* is a function of mean and Figure 1. The effect of insurance on the use
variance responses to input use and of risk aver- of a risk-increasing input (13 = 0.1269)
sion. Let a = /xx/I, 13= xx/uT, s = t/pI, R
= pA and p = Rs2/2. a and P are the elasticities
of output mean and variance with respect to in- S
.3532, = .7064
put application. s is the coefficient of variation
of output and R is the coefficient of relative risk
aversion evaluated at the mean level of output. Dc.
p is a second order approximationto the relative
risk premium, i.e., the fraction of mean income
that the producer is willing to pay as risk pre-
mium (Newbery and Stiglitz, p. 73). Then
0o
(12) x(r) 4 x(O) as r* = 2 - a/(2pp).
r-
The critical level of insuranceis higher for larger
values of variance elasticity and risk premium
0.2 0.3
and for smaller values of the mean elasticity. 0.0 0.1 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Insurance

9 The derivation of equations (9) and (10) is available from the Figure 2. The effect of insurance on the use
author. of a risk-increasing input (f. = 0.7064)
920 November 1993 Amer. J. Agr. Econ.

input use at all levels of r. Figure 2 is calculated in two stages. In the first stage, define a cost
using the same mean elasticity as in figure 1 but function C(Qp,a) by
the variance elasticity is twice the mean elastic-
(14) C(p, a) = mini wx, i = 1, ..., n
ity. As a result, the risk reduction effect is
stronger than the moral hazard effect for levels subject to x E V(Ap,a)
of insurance up to r* = 0.7. Input use is first
where V(pt, a) = {x: t(x) ?!t, x) < or}.Be-
increasing and then decreasing in r. cause C(Q, a) is like a multi-output cost func-
Thus, the numerical simulation suggests that
the risk reduction effect is stronger (and the im- tion, it is well defined and differentiable if the
set V is nonempty, closed, and convex (Cham-
pact of insurance on expected supply positive) bers 1988, ch 7). It is also clear, from standard
if insurance is limited, if the producer is highly
risk averse, and if the impact of input use on elementary arguments, that C, > 0 and C, < 0.
In the second stage, a producerchooses A and
output variability is much larger relative to its a to solve
impact on expected output.
r(t, o, I, P) = EU(Q(pt, o; 0)
(15)
max,•,,
+ I(Q(, aor))- P - C(Q, a))
Insurance and Supply Response: where Q(A ,oa;0) = A + ?O. Clearly, Q(t, oa;
A Multiple-Action Model 0) = q(x, 0) whenever x E V(pt, a).
The two-stage problem in (14) and (15) can
For reasons mentioned earlier, a straightforward be considered as a dual to the problem in (13).
extension of the single-input case to multiple in- In the dual of the certainty case, a producer
puts is not fruitful. In this section, I consider a chooses output subject to a cost of choosing of
dual approach to the problem in which a pro- that output. Here in the uncertainty case, a pro-
ducer directly chooses the parametersof the out- ducer chooses an output distribution subject to
put distribution. Suppose a production function the costs of choosing that distribution. It re-
is of the form q(x, 0) = gt(x) + o(x)0, where mains to be shown that the primal and dual for-
0 is a zero mean and unit variancerandomshock, mulations solve the same problem. To see this,
and x a vector of n inputs. Such production it is enough to show that the maximized ex-
functions, first proposed by Just and Pope, have pected utility from (15) is not less than the ex-
been widely used in theoretical and empirical pected utility associated with any arbitraryinput
work. Just and Pope showed functions of this vector k. Let ^ = t(i)k)and 60 = o(•). Also let
type to be less restrictive than the popular al- A*, o*, and x* be the solution to (15) and (14).
Then

EU(q(k, 0) + I(q(k, 0)) - P - wi) = EU(tt(i) + + I()t(A) + o(ri)0) - P - w )

= EU(2 + 60 o()0? + I(42 + 6) - P - wi)
EU(2? + 60 + I(P + 60) - P - C(P, 6))
=- EU(Q(4, &; 0) + I(Q(, ; 0)) - P - C(^, &))
? EU(Q(p*, o*, I(Q(,0*,
0) + 0)) - P - C(,*, o*))
= EU(q(x*, 0) + I(q(x*, 0) - Por*; - wx*)
Hence the solution from the two stage maximi-
ternative of specifying a multiplicative produc- zation is identical to the solution from (13).
tion shock.1o The primal approach consists of Let g(I, P) and (0O,0) be the mean outputs
solving the following program. when producers optimally choose inputs in the
presence and absence of insurance respectively.
(13) maxx, EU(g4(x) + o(x)0 + I(gt(x) The question again is how purchaseof insurance
+ o(x)0) - P - wx), i = 1,..., n alters mean supply. Solving (15), the mean out-
where w is a vector of n input prices. puts with and without insurance can be directly
compared. The comparative statics are carried
Consider an alternative problem to be solved out
by assuming F(A4,a, 0, 0) to satisfy the suf-
ficient conditions for concavity, namely, #, <
10The dual
approach is also available for the multiplicative case. 0, 4q, < 0 and If,,f - > 0. In order to
Suppose q = g(x)O. The dual in this case considers the choice of ensure strictly positive solutions, it is also as-
A alone and the problem reduces to the selection of one variable, sumed that lim,,0 = 0 and lim0o
which can then be handled by slight modification of the single input
case considered earlier (see, for example, Eeckhoudt and Hansen).
Co(/, o) C(/i,
o)=
Ramaswami Supply Response to Insurance 921

No-Insurance Case A sufficient condition for determining the rela-

tion between 0oand p, is stated below.
In the initial situation, I(q) = P = 0 for all q
and the marginal changes in expected utility are PROPOSITION 3: The expected output of a risk
averse producer is greater than (equal to, less
given by
than) the expected output of a risk-neutral pro-
(16) or, ,0) ducer if C,. > (=, <) 0.
q•i,(/,= 0, 0))(1 - C,(/, a))
EU'(ir(/p, o, PROOF: Define (g) = ,(u, a(g), 0, 0). Then
(17) qi,(f4, o, 0, 0) 0'(g) = ,i Ig+ + But applying the im-
O'a'(/Lg).
= 0, 0))(0 - Co(t/, o)) plicit function theorem to (22), Oa',() = - (I',i
EU'(Trr(/,o, /1i,,). Substituting, 0'(g/) = (4,Ii ,i -- 2)/
where wi(u, o, 0, 0) = Q(,p, o; 0) - C(,p, oa).
qi,, < 0 by our assumptions on 4. Since 4Pis
Rewriting the above equations, decreasing in g and = 0, ~< n as
g0
(p0o) Oa('4n), (•4(n)
0. Now 0, 0)) = 1
(18) iqf,(po,u, 0, 0)/EU'(Trr(u,o, 0, 0)) - 44(1ni)/EU'(OT(gn,
= 1 - q,(u), a) and - C, (/,n, a(/,n)).
=
Using (21),
-0(nn)/EU'(Qr(gLn,
Ua(/gn), 0, 0)) CQ(/n, m,(gn)) Oa(Qn)).
But > Oa(/j) for all g. Hence,C,(/Ln,
(19) qiju,o( ,O, 0)/EU'(rr(,p, a, 0, 0)) 'm(/g)
= - 4(A, a) + 0, 0)) 0. Be-
0 if C,, - 0(nn)/
EU'(Tr(/,n, Oa(/Ln),
cause is
cov(U'(rr(,u, o, 0, 0)), 0)/EU'(ir(/u, o, 0, 0)) expected marginal utility strictly posi-
tive, (pn) 0 if C,,- 0.
For a risk-neutral producer,
(20) ,(i , o, 0, 0) =
- C,(,p, a)[EU'(ir(,p, o, 0, 0))] > 0 for all or. Risk Reduction and Moral Hazard Effects
Hence for a given u, a risk-neutral producer With insurance, the equations analogous to (18)
and (19) are

(24) f,(,L, a, I,P)/EU'(rr(,p, a, I, P)) = 1 - C,(p, a)

+ EU'(Trr(Q,
a, I, P))I'(q)/EU'(ir(,p, a, I, P))
(25) i,(, u, I, P)/EU'(ir(pu, a, I, P)) = - C, (u, a)
+ cov(U'(ir(,p, a, I, P)) (1 + I'(Q)), a, I, P))
0)/EU'(rT(,u,
where
chooses the maximum possible or. Let om(/L)
=
max o-(x) subject to u(x) = u. Substituting in 1T(,/, a, I, P) = Q(/L,a; 0)
(16), a risk-neutral producer's choice of mean + I(Q(gp, 0)) - P - C(Qp,a).
ur;
output, satisfies Let be an insuredproducer'soptimal choice
g,n, o'i(p)
(21) of a as a function of u. It satisfies
Ii(Vp,,, 0, ,,m(o, n), 0, 0))
jm(/(Ln),
= (1 - (26) = 0
0)/EU'(T(/J,n,
C,(/,n, -Co,(g, oai(g)) + cov(U,'(7r(Q,
.m(/n,,))
Uoi(g), I, P))(1
Let be a function which describes a risk + I'(Q)), I, P)) = 0
averseO'a() 0)/EU'(r(tp, oi(p),
producer'schoice of orfor a given u. o'a(p) Denote u P). ui solves
solves ui p(I,

= (27) 1 - C,(,ui, oi(,i)) + EU'(Tr(,ui,oi(/i),

(22) fi,(/L, a(/), 0, 0)
0, 0))(0 - = 0. I, P))I'(Q)/EU'(r(pui, oi(,i), I, P)) = 0.
EU'(Irr(/, ua(/.L), C,(/L, Oa(/g.)))
Because qif < 0, (20) and (22) imply < Comparing the first order conditions for b/ with
for all u. in O'a(g)
the first and without insurance, that is, equations (23)
Om(/L) Substituting O'a() and (27), note that with o held constant, insur-
order condition for g, a risk averse producer's
choice of mean ance decreases expected output because of the
optimal output, g0, solves thirdterm in (27), which is nothing but the moral
(23) hazard effect. However, insurance changes the
I1,(/Lo,
ua(L0o), 0, 0)/EU'(T(/Lo, a(rQ), 0, 0)) choice of o and this also matters in determining
= 1- C
,(/LO,Oa(ULO))the final impact of insurance.
922 November 1993 Amer. J. Agr. Econ.

From the proof of Proposition 3, we know (P (Ahsan, Ali and Kurian, Nelson and Loehman),
is decreasing in u and = 0. Then, p~i the moral hazard effect is absent because insur-
as 0. Now 0(p.o) ance indemnities are contingent on input levels.
,ro 0(,ji)/EU'(1T(;,li,
0, 0)) =(•.•i)
1 - C, (Li,, a(Wui)).Using (27), 'a(,i),
(28) 0, 0)) = C oPi()Li))
-
CL' (,i, o(.(Ii))
•(,i)/EU'(iT(uik, iOa(/Li), (Li),i,
- i, Qri(tui),IP))I'(q)/EU'(iTr(piio-i(t).,lP)).
EU'(iTr(Q
The last term, which is the moral hazard effect, If the risk reduction effect is the sole impact,
is positive and by itself (i.e., ignoring the first
mean outputadjustsin the directionof risk-neutral
two terms) reduces mean output. The difference
levels. In the presence of moral hazard, how-
between the first two terms is the risk reduction
effect. The sign and magnitude of the risk re- ever, this need not necessarily happen. The nu-
merical simulation suggests that the risk reduc-
duction effect depends on the difference be- tion effect dominates moral hazard only if risk
tween and Oa(,pi) and on the sign of the aversion is high and if the impact of input use
o'(,ui)
cross derivative
C,,, on output risk is large relative to its impact on
4: For all constant and decreas-
PROPOSITION expected output.
ing risk averse utility functions and for convex Future work could extend these results to al-
insurance schedules, > Oa(j) for all tLand low activity choice. It is often suggested that
oi(,u) crop insurance leads to cultivation of riskier
crops, adoption of riskier production tech-
(29) C,(/, o0(7i)) - C,(;, Oa(Ui)) niques, and use of marginal,high-riskfarm lands
> 0 ifC, > 0 and
(see for example, Todd, Gardner, and Kramer).
(30) - C (
In other words, crop insurance could increase
C,((i/, Oi(Li)) 0i, 0a(•Li))
< 0 ifCl < 0. expected supply by promoting specialization in
the riskier activities. Such argument can be ex-
Proposition 4 is proved in the appendix. Its amined in the dual formulation proposed here
implication is that insurance decreases expected by interpretingx as a vector of all producer de-
output if C,, > 0. On the other hand, if C1, < cisions (including activity and input choices) and
0, the moral hazard and risk reduction effects q as aggregate revenue from all activities. Re-
run in opposite directions and the net effect is sults obtained here suggest that insurance may
indeterminate. Note that the class of convex in- not always increase expected supply. This would
surance schemes includes real-world insurance not be surprising, since it seems reasonable to
schedules which are piecewise linear. suppose that moral hazard in input choice might
limit the specialization achievable by insurance.
Risk averse farmers would specialize in the risk-
ConcludingRemarks ier activity only if they were to be offered suf-
ficient insurance. But that may reduce input use
I have decomposed the impact of crop insurance in the riskier activity due to the moral hazard
on variable input use into a moral hazard and a effect. If input use decreases, expected returns
risk reductioneffect. The principalinsight is that from the riskier activity also decrease, which
insurance changes the marginal costs of input limits specialization in the riskier activity.
use in two ways. First, insurance reduces risk These findings point to a trade-off in the de-
and therefore reduces the wedge between ex- sign of insurance. In many developing coun-
pected marginal product and input price due to tries, agricultural insurance programs are ex-
risk aversion. This is the risk reduction effect pected to increase agricultural production by
which leads risk averse decisions toward risk- inducing the use of riskier inputs and technol-
neutral levels. Mean output increases or de- ogies. The implicit belief is that insurance pro-
creases depending on the underlying technol- grams have strong risk-reduction effects. How-
ogy. Second, insurance reduces the marginal ever, this is true only in the absence of moral
productivity of all inputs, as an increase in out- hazard. In the real world, insurers find it in-
put is always accompanied by a decrease in ex- feasible to monitor all production activities of
pected insurance indemnities. This is the moral farmers. A badly designed insurance program
hazard effect which reduces the use of all inputs could have moral hazard effects which com-
and decreases mean output. In earlier work pletely erode the incentives to produce higher
Ramaswami Supply Response to Insurance 923

output due to risk reduction. One way to control ed. T. Bewley, Cambridge University Press, 1987.
moral hazard is to limit the amount of insurance Holmstrom, B. "Moral Hazard and Observability." Bell J.
through higher deductibles or lower price elec- Econ. 10(Spring 1979):74-91.
tions. Unfortunately, lower amounts of insur- Just, R. E., and R. D. Pope. "Production Function Esti-
ance also have correspondingly weaker risk mation and Related Risk Considerations."Amer. J. Agr.
Econ. 61(May 1979):277-84.
reduction effects. Similarly, while insurance
schemes contingent on area rather than individ- Kaylen, M. S., E. T. Loehman, and P. V. Preckel. "Farm-
level Analysis of Agricultural Insurance: A Mathe-
ual yield can reduce and even eliminate moral matical Programming Approach." Agricultural Sys-
hazard, their risk reduction effects on individual tems 30(1989):235-44.
producers are also smaller. The challenge, Kramer, R. "Federal Crop Insurance: 1938-1982," Agri-
therefore, is to design and administer insurance cultural History 57(1983):181-200.
programs which reduce moral hazard while pre- Leathers, H. D., and J. C. Quiggin. "Interactionsbetween
serving risk reduction effects. In evaluating al- Agricultural and Resource Policy: The Importance of
ternative insurance schemes, simulation exer- Attitudes toward Risk." Amer.J. Agr. Econ. 73(August
cises could be used to assess the trade-offbetween 1991):757-64.
risk reduction and moral hazard effects. Lewis, T., and D. Nickerson. "Self Insurance against Nat-
ural Disasters." J. Environ. Econ. and Manag.
[Received January 1992; final revision 16(1989):209-23.
received February 1993.] Lloyd, A. G., and R. G. Mauldon. "AgriculturalInstability
and Alternative Government Policies: The Australian
Experience." Crop Insurance for Agricultural Devel-
opment, ed. P. Hazell, C. Pomareda and A. Valdes.
Baltimore and London: The Johns Hopkins University
References Press, 1986, pp. 156-77.
MacMinn, R., and A. Holtmann. "Technological Uncer-
Ahsan, S. M., S. G. Ali, and N. Kurian. "Towards a The- tainty and the Theory of the Firm." South. Econ. J.
ory of Agricultural Insurance." Amer. J. Agr. Econ. 50(1983):120-36.
64(August 1982):520-29. Mirrlees, J. "The Theory of Moral Hazard and Unobserv-
Antle, J. M. "Pesticide Policy, Production Risk and Pro- able Behavior-Part I," Mimeo, Nuffield College,
ducer Welfare: An Econometric Approach to Applied Oxford, 1975.
Welfare Economics." Resources for the Future. Nelson, C., and E. T. Loehman. "FurtherToward a Theory
Chambers, R. "Applied Production Analysis," Cambridge of Agricultural Insurance." Amer. J. Agr. Econ.
University Press, 1988. 69(August 1987):523-31.
. "Insurability and Moral Hazard in Agricultural In- Newbery, D. M. G., and J. E. Stiglitz. The Theory of
surance Markets." Amer. J. Agr. Econ. 71(1989):604- Commodity Price Stabilization,Oxford University Press,
16. 1981.
Eeckhoudt, L., and P. Hansen. "Minimum prices and Op- Olson, K. D., and V. R. Eidman. "A Farmer's Choice of
timal Production under Multiple Sources of Risk: A Weed Control Method and the Impacts of Policy and
Note." Eur. Rev. Agr. Econ. 16(1989):411-18. Risk." Rev. Agr. Econ. 14(1992):125-37.
Gardner, B., and R. Kramer. "Experience with Crop In- Pope, R. D., and R. Kramer. "Production Uncertainty and
surance Programs in the United States." Crop Insur- Factor Demands for the Competitive Firm." South.
ance for Agricultural Development, ed. P. Hazell, C. Econ. J. 46(1979):489-501.
Pomareda and A. Valdes. Baltimore and London: The Ramaswami, B. "Production Risk and Optimal Input De-
Johns Hopkins University Press, 1986. cisions." Amer. J. Agr. Econ. 74(November 1992):860-
Hazell, P., M. Bassoco, and G. Arcia. "A Model for Eval- 69.
uating Farmers' Demand for Insurance: Applications Ramaswami, B., and T. L. Roe. "Crop Insurance in In-
in Mexico and Panama." Crop Insurance for Agricul- complete Markets." Contributions to Insurance Eco-
tural Development, ed. P. Hazell, C. Pomareda and nomics, ed. G. Dionne, Kluwer Academic Publishers,
A. Valdes. Baltimore and London: The Johns Hopkins Boston, 1992.
University Press, 1986. Ray, P. K. Agricultural Insurance: Theory and Practice
Hazell, P., C. Pomareda, and A. Valdes. "Introduction." and Application to Developing Countries, 2nd Edition,
Crop Insurance for Agricultural Development, ed. P. Pergamon Press, Oxford, 1981.
Hazell, C. Pomareda and A. Valdes. Baltimore and Rothschild, M., and J. E. Stiglitz. "Increasing Risk I: A
London: The Johns Hopkins University Press, 1986. Definition." J. Econ. Theory 2(September 1970):225-
Hildreth, C., and L. Tesfatsion. "A Note on Dependence 43
between a Venture and Current Prospect." J. Econ. Todd, R. "ExpandedFederal Crop Insurance:A Better Way
Theory 15(August 1977):381-91. for Taxpayers to Share Farmer's Risks." Federal Re-
Hart, O., and B. Holmstrom. "The Theory of Contracts." serve Bank of Minneapolis Quarterly Review Win-
Advances in Economic Theory, Fifth World Congress, ter(1982): 18-28.
924 November 1993 Amer. J. Agr. Econ.

Appendix Hence

(A3)
Proof of Proposition I
G(x,, G(x,,
0 0)dF(0)> 0*") 0)dF(0).
Substituting for caMu,aM and p, qx,(xi,
For all 0 >0)q,(x•,
0*, q,(x,, 0) > qj(x,, 0*). But also for these
(A1) oa,(xi, I, P)p(x,, I, P)oa,(x,) values for 0, G(x,, 0) > 0 (by lemma). Hence
= -cov(U'(rT(xj, I, P)), q,(x,, 0))/
0)]. (A4)
[EU'(T(xj,l,P))Eu(xi, OM OM
The sign of (Al) is opposite to the sign of cov(U'(rr(x,, I, >
G(x,, O)q.(xi, 0)dF(0) q,(x(, 0*) G(x,, 0)dF(0)
P)), 0)). U' is decreasing in 0 since dU'/dqO = U'(rr(x,
qx(x,,
I, P))(1 + I'(q))qo < 0 given the assumptions q0 > 0, I'(q)
> -1 and U' < 0. On the other hand q, is monotonic in- Combining (A3) and (A4),
creasing or decreasing in 0 depending on the technology.
Thus ,Mu(x,,I, P)p(x,, I, P)oMp(x,) is positive if the input cov(G(x, 0), q(x, 0)) > q,(x(,0*") G(x,,0)dF(0)
is risk increasing and negative if the input is risk decreas-
= =
ing. This proves the first inequality in (i) and (ii). q,(x(, 0*)EG(x,, 0) 0.
Next, consider the relationship between OM(x,, I, P)p(x,
Let D =
This proves the second inequality in (i). The second in-
I, P)aoM(xi) and oaM(x,, 0, O)p(x,, 0, O)ap(xi).
- equality in (ii) is proved similarly as well.
cMU(xi,0, O)p(x,,0, 0)aMp(x) oUM(Xi,I, P)p(x,, I, P)oM(x,). Turning to the proof of the lemma, note that
The proof consists in showing D to be positive (or negative)
if x is risk-increasing (or risk-decreasing). Substituting for G(x,, 0) - 0 as U'(7Tr)/U'(Tro) -
and p, EU'(Tr)/EU'( ro).
OMU, PMp
Let T(O) = Now T, = i7T0 + (x,, 0) where
D= v = I(q) - P. U'(i'T)/U'('o).
Since {I, P} is a feasible contract, P = EI(q(x,
-cov(U'(r(x;, 0, O)),q_(x;,0))/EU'(rr(xi,O,O)E,,(x,, 0) 0)) or Ev(x,, 0) = 0. Further, as I is decreasing in 0, v is
+ cov(U'(ir(x,, I, P)), q_(x(,0))/EU'(ir(x,, I, P))E,(x,, 0). also decreasing in 0. Thus there exists an unique 0 such
that I(q(x,, 0)) - P or v(x,, 0) > 0 for all 0 00. From the
Let rr(x,, I, P) = irr and rr(xi, 0, 0) = iro. Multiplying
monotonicity of U' it follows that
throughout by Eq,,
(A5) T(0) > 1 for all 0 > 0 and
D[E,(xi, 0)] = cov(U'(ir), q,(x,, 0))/EU'(i,;) T(0) - 1 for all 0 0.
- cov(U'(Tro),
qx(xi,0))/EU'(iro) Let C =
-
Because Ev(x,, 0) = 0 and v
= (EU'(Tr,)q(x,, 0) - EU'(iT)/EU'(Tro).
EU'(Tr)Eq(x,, is decreasing in 0, 7r0ois riskier than i;, by a mean preserving
- (EU'(Tro)q1(x,, 0))/EU'(t1i)
0) - EU'(iTo)E,(x,, 0))/ spread. But if risk aversion is non-increasing, marginal util-
= EU'(i)q,(x,, 0)
EU'(1ro) /EU'(rO,) ity is convex. It follows (Rothschildand Stiglitz) that EU'(IT,)
- EU'(7To)qx(x,, 0)/EU'(iro) < EU'(iro) and C < 1. Let 0* be such that T(0*) = C. 0*
= E{U'(iT)/EU'(T,) - 0) exists because G(x,, 0) ' 0 as T(0) ' C and EG(x,, 0) =
U'(Tro)/EU'(Qro)}q,(xx,
= E(G(x,, 0)qj(x,, 0)) 0. Due to (A5) and the fact that C < 1, 0* must be less
where G = U'(iri)/EU'(Ti,) - U'(iT)/EU'(iT). But EG(x,, 0) than 0. But T is increasing in 0 in this range. Hence 0* is
= 0. So unique.
(0) -
(A2) D[Eq,(xi, 0)] = cov(G(x,, 0), q(x,(, 0)) Now, T'(0)
U'(Tro)U"(IT,)IT: U'(Tr,)U"(Tro)TrO(0)
=Uo)
The following result, proved later, is useful in signing the -
above covariance. U'(io)U"(iT)qo(1 + I'(q)) U'(ir)U"(iro)qo
U'(,ro)2
LEMMA: For all non-increasing risk averse utility func- {U' - U' (i,)U"(iTo)}qo
tions, there exists a 0* such that G(xi, 0) > 0 for all 0 > ("o)U"(Wi)
U'(ITo)2
0* and G(x,, 0) < 0 for all 0 < 0*.
U'(iro)U"(iT,)'(q)qo
(A2) can now be signed using the lemma and the meth- U'(7To)2
ods of Hildreth and Tesfatsion. If F is the cumulative den-
sity of 0 with support [0,, 0,M], (A2) can be written as (A6)

T'(0) = )
- + j (
]
cov(G((x,,0), q,(x,, 0)) A(j,))
U(T)J(A(] 0)
= G(x,0), 0)dF(0) + G(x,, 0)dF(0). where A(IT) is the coefficient of absolute risk aversion eval-
0)(x;, O)q,(x,,
uated at 7T.The second term in the above expression is pos-
Suppose > 0. Then for all 0 < 0*, q,(x,, 0) < q,(x, itive since U" < 0 and I'(q) O. The first term is non-
-
qo for these values for 0, G(x,, 0) < 0 (by lemma).
0*). But also negative for non-increasing risk averse utility functions
Ramaswami Supply Response to Insurance 925

whenever iTr > 1ro,i.e., 0 < 0. Thus T is increasing in 0 (A7) 0,(A, o, 0, 0)/EU'(Tr(Q, o;, 0, 0)) =
for all 0 - -0. -cov(U'(Tr(tL, oi, I, P))(1
+ I'(q)), 0)/EU'(7r(Ap,o, I, PP))
Because T'(0) > 0 for all 0 < 0,
+ cov(U'(ir(p, 0, 0)), 0)/EU'(7r(A, o;, 0, 0))
T(O) < C for all 0 < 0* and o';,
C for all 0 E [0*, 0]. = cov(U'(ir(A, o, 0, 0)), 0)/EU'(Tr(Q, 0, 0))
T(0) - o',
- cov(U'(rr(Ap,o, I, P)), 0)/EU'(7r(tL, o, I, PP))
But from (A5), T(0) > 1 > C for all 0 > 0. Hence, 0* -
cov(U'(r(iT , o-,I, P))I'(q), 0)/EU'(Tr(Q, o, I, PP)).
satisfies
The difference between the first two terms can be shown
T(0) < C for all 0 < 0* and to be negative by the methods used to prove part (i) of prop-
T(0) - C for all 0 > 0*. osition 1. The details are not repeated here. What remains
This proves the lemma. to be shown is that the third term is also negative. This
term is negative if U'(7r(AL,
o-, I, P)I'(q) is increasing in 0.
PROOF OFPROPOSITION 4: (29) and (30) follow trivially once Consider 0, > 02. Let U', I' and U', I' be the marginal
it is shown that 0o(L) > o(pj). For notational simplicity, utility and slope of the insurance schedule evaluated at 0,
the functional dependence of o, and oaon A will henceforth and 02 respectively. Then we need to show U'I' > U' I.
be suppressed. Because I,(AL,o-, 0, 0) = 0 and , Marginal utility U' is decreasing in 0 since aU'/a0 = U'(1
qf,,(I o,
0, 0) < 0, oa > oa if fIq(Q, o, 0, 0) < 0. Now 0,(A(, o;, + I')o < 0. Consequently, U' < U'. Since I1 < 0, U' I' >
0, 0)/EU'(r(T , 0, 0)) = -C(L, o')) + cov(U'(Ir(TQ,o', U'I'. On the other hand, I' > I1 by the convexity of I. But
o',
0, 0)), 0)/EU'(rr(, 0, 0)). U' > 0 and so U'I' > U' I. Combining the inequalities,
Substituting for C(Q,o';, (from (26)), UlI > u'I'.
oi)