1 and the fact that p*(y) = l/(1 - S) y, when )3.= 1, it follows that lim,,, 6’Ep(y’) y’ = 0, and the limit policy is optimal. 9. Hence lim,,, fYEp(y’)y’ < lim, +oo6’ max(C/l - &p(a)y^] = 0, which concludes the proof. I COROLLARY 3.3.4. If _uhas absolute risk aversion, R, > K, K > 0, then the limit policy is optimal. /?-a. Similarly S + A > j?. Since J?(Z)= Ep(z + w), J?is constant in [s, v] and p is non-increasing, p(y) has to be constant in [s + A, v + A]. As before, we now conclude that A - a > s - v = r(j3 - a) and so integrating obtain A - a > r”(j3 - a) for all _n,a contradiction.
ON INCOME FLUCTUATIONS
AND CAPITAL GAINS
31
(2) Let us suppose0 > e,, > -1. Using Lemma 3.3.2 for 0 < nl < 1, such that 6r’-’ < l? and Theorem 3.3.1 for u*‘(y) = l/y’ we get p,(y)
3.3.3.
If u(y)
is bounded above then the limit
policy
is
optimal. Proof. From the concavity and differentiability of u we have that given any y’, for all y > y’ U’(Y) Y < -U(Y’> + 4~) + U’(Y)Y’. Since lim,,, u’(y) = 0 we can write, for a given e > 0 and starting from a certain y^ that u’(y)y < cy’ + (K - u’(y)) = C > 0, where K is such that u(y)
Pro& From R, = -u”(y)/u’(y) > K we can write u”(y)egy f Ku’(y) eKy < 0, that is, u’(y) eKy is strictly decreasing. Using that lim,,, u’(y) eRy= M for some M> 0 and the fact that _uis increasing, we get that u(y) is bounded above and then, by Theorem 3.3.3 the limit policy is optimal. I THEOREM
3.3.5.
If u’(y) y is decreasing then the limit policy is optimal.
ProoJ Given a y*, for all y 2 y* we have that u’(y) < Cly, where c = u’(y*) y*. Repeating the argument used in the proof of Theorem 3.3.3 we get the desired result. m An example of an utility function with a zero asymptotic exponent and no optimal policy. Let K > 0. u(y)= 1 -ee-‘+K.p, Suppose that there exists an optimal policy. Then p(y) > e pcfy) + K, for all y. Hence hm,,, p(y) > K > 0, a contradiction by Proposition 2.3. The following result holds generally, since p(y) > u’(c(y)). PROPQSTION
642132/l-3
3.3.6.
Iflim,,,
uf (y) > 0, them there is no optirna~p~l~~~l.
32
MARILDA
A. DE OLIVEIRA SOTOMAYOR
4. SOME RESULTS ON A CAPITAL FLUCTUATION 0<6< l,r> 1,6r=l
PROBLEM:
(A) Deterministic Case: w E a. LEMMA
4.1.
xy( y) = 0, for all y < a and for all t.
Proof. Suppose that there exists a f such that x:(a) > 0. Then p;(a) < py(rxy(a) + a) and thus a > rxy(a) + a > a, a contradiction. Hence x:(y) = 0 for all y < a. I LEMMA
4.2.
cT(y) > 0 for all y > 0 and for all t.
LEMMA
4.3.
xf( y) > 0 for all y > a and for all t.
LEMMA
4.4. x”(y)
> 0 for all y > a.
The proofs of the three last lemmas are, as that for Lemma 4.1, immediate. THEOREM 4.1. For the limit policy we have yLtl < yi, ‘dt, if y” > a and yi = a, tit > 0, ify” < a.
Proof. THEOREM
It is immediate from Lemma 4.3. 4.2.
1
The limit policy is optimal.
ProoJ It suffices to show that lim,,, #p”(yi) y: = 0, since the limit policy is competitive with p”(y) = lim f+o3pF( y). Using Theorem 4.1 this fact obviously follows. m Now, we will see that there exists only one optimal policy for all utility function. Let x,(y)==,
r
= 0,
if
y>a
if
O
and c
(y)=
(r-
a
l)y+a
r =Y,
’
if
y>a
if
O
ONINCOMEFLUCTUATIONS
ANDCAPITALGAINS
33
SettingP,(Y) = u’(c,(Y>> we have that the given policy is optimal, no matter the utility function. Using Theorem 4.2 and the unicity of the optimal policy when it is the limit policy we get that the given policy is the limit policy and the unique optimal one. (B) Stochastic
Case, i.e., w E [a,A]
is a Non-degenerate Random
Variable THEOREM
4.3.
The limit policy is optimal.
ProojI It suffices to show (1.7) with p(y) = lim,,,p,(y) and so to use Proposition 1.3. r we deduce, respectively, that Ep(y’) > From (1.6) and x(y)
4.4. p(y)
is a strictly
decreasing function,
where p(y) =
lim,+,p,(~). Proof. (1) P(Y) = U’(4Y))
Suppose p-‘(M) = [q/3], 0 < a 3 < co, thus from we have c(y) =K and x(y) = y -K for y E [q/3]. Let $) =L~E$(;~+~~)
(2) If p-I(M)= [ a, co), then, as before, x(y) = y -X for all y > a, contradicting the fact that x(y) Q x0(y) = y/r, r > I, for y > a. THEOREM
4.5.
P(lim,,,yf
= co) = 1.
ProojI As in the proof of Theorem 2.8, we have that Z, = -p( y’) is a semi-Martingale with an almost sure limit. By Theorem 4.4 p- ’ exists and is continuous. Hence y’ =p-I(-Z,) converges almost surely.-Since the set of sample paths for which there are infinitely many _t such that wf = A and wtt ’ = a has probability one we can assumethat the sample paths on which y’ converges also have this property. This fact along with the continuity of x(y) and the relation yff ’ = rx(y’) + W( contradicts a finite limit for yf In case 6r < 1, under certain condictions there exists Y; such that, for all y > Y; rx(y) + A < y, implying yf > yff ’ if yf 27. In case 6r > 1, under certain condictions, too, there exists a level of
34
MARILDAA.DE
OLIVEIRA SOTOMAYOR
capital, F, such that for all y > 7, rx(y) + a > y, and hence y’ < y’+ ’ if y’>y. In our case, 6r = 1, we have that. y
for ally >A.
y>rx(y)+a>Y-(A-a), Although yf --f co when t--f co almost surely, there is no capital level starting from which the optimal sequence of capitals is increasing, since the only way for this to occur is to have y = rx( y) + a, starting from a certain y, which cannot happen, as the following proposition shows. PROPOSITION
4.6.
There is no L such that for all y > F, rx( y) + a = y.
Proof. If there will exist a such F we would have p(rx( y) + a) =p( y) = Ep(rx( y) + w) implying that p(y) = p( $) > 0, for all y > J, a contradiction
by Theorem 4.4. Remark. If u’(y) is convex then { y”} is a semi-Martingale since p(y) is strictly convex and p( y’) > p(Ey’+ ‘). Theorem 4.5 corrects the erroneous result of [7] for the case 6r > 1, where an attempt was done to prove, using u(y) = -e-y, that by an appropriate choice of the random variable w and of its distribution the optimal sequence of capitals { yf}t is bounded almost surely. We found in Section 3 the limit policy for u(y) = -ePy and any w.
x(Y)=:+
--&
lg Ee-(‘-W’)W
lg Ee-
+ (r-
1 r-l
1/r)w
. lg 6r.
----&lg
6r.
ACKNOWLEDGMENTS I would like to express the subject developed here Svetlichny, for his advising de Araujo, for his interest This work is part of my was financially supported Tecnologico) and CAPES of the Brazilian government.
my gratitude to Professor Jack Schechtman, who introduced me to and for his valuable assistence and advising; to Professor George and patient and instructive comments; to Professor Aloi’sio Pessoa and suggestions. thesis presented to the Department of Mathematics of PUC/RJ and by CNPq (Conselho National de Desenvolvimento Cientifico e (Coordenapao de Aperfeicoamento de Pessoal de Nivel Superior)
ON INCOME FLUCTUATIONS
AND CAPITAL GAINS
35
REFERENCES 1. J. L. DOOB, “Stochastic Processes,” pp. 324-325, Wiley, New York 1953. 2. E. KOHLBERG, A model of economic growth with altruism between generations, J. Econ. Theory 13 (1976), I-13. 3. D. LEVHARI, L. J. MIRMAN, AND I. ZILCHA, Capital accumulation under uncertainty, hternal. Econ. Rev. 21 (1980) 661-671. 4. D. LEVHARI AND T. N. SRINIVASAN, Optimal savings under uncertainty, Rev. ECGPZ. Studies 36 (1969), 998-1019. 5. J. SCHECHTMAN, Some applications of competitive prices to dynamic programing problems under uncertainty, ORC 73-5, Operations Research Center, University of California. Berkeley, March 1973. 6. J. SCHECHTMAN, An income fluctuation problem, J. &on. Theory 12 (1976), 218-241. 7. J. SCHECHTMAN, A grain storage problem, with random production, ORC 74-3, Operations Research Center, University of California, Berkeley, January 1977. 8. J. SCHECHTMAN AND V. L. X. ESCUDERO,Some results on an income fluctuation probiem, J. Econ. Theory 16 (1977), 151-166.