Rawls's Principle of Just Saving Author(s): Kenneth J. Arrow Reviewed work(s): Source: The Swedish Journal of Economics, Vol.

75, No. 4 (Dec., 1973), pp. 323-335 Published by: Blackwell Publishing on behalf of The Scandinavian Journal of Economics Stable URL: http://www.jstor.org/stable/3439143 . Accessed: 04/02/2012 21:08
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Blackwell Publishing and The Scandinavian Journal of Economics are collaborating with JSTOR to digitize, preserve and extend access to The Swedish Journal of Economics.

http://www.jstor.org

OF RAWLS'S PRINCIPLE JUSTSAVING*

KennethJ. Arrow
Harvard University, Cambridge, Massachusetts, USA

1. The Problemand the Model

The problem of justice in the distribution of resources is of course basic. At no time in the history of economic thinking has there been a thoroughly agreedon criterion, but at least among the more philosophic circles of economists a utilitarian criterion has been more or less accepted. That is, we assume there exists for each individual i an interpersonally meaningful cardinal utility function, U7(c), where c, is the consumption of the ith individual; then a just or optimal distribution is one which maximizes,

U(c,)
where the variables c, are constrained by technology, initial resources, and informational and other difficulties in transferring goods among individuals. Rawls [1971] has proposed an alternative criterion for just allocation. He imagines all the individuals in the society getting together in an "original position" where each knows the possible resource allocations but no one knows which member of the society he is going to be. They must decide on a criterion function for the distribution of goods. Then, he argues, they will find it rational to agree not on the sum-of-utilities criterion but on the rule of maximizing the welfare of the least well-off member of the society (since this might be any one of them), i.e., they maximize,

min Uj(ci)
subject to the same constraints on the variables c,. I ignore here the richness of Rawls's discussion, some other constraints he imposes on the allocation of resources (particularlysetting an infinitely higher value on liberty than on goods),
* This work was supported in part by National Science Foundation Grant GS-3269-A2 to the Institute for Mathematical Studies in the Social Sciences, Stanford University. This paper was begun at Churchill College, during the author's tenure of a John Simon Guggenheim Memorial Fellowship, and completed at Stanford University. I am indebted to Partha Dasgupta, discussions with whom focused my diffuse interest in Rawls's theory of just saving into seeking an exact characterization, to Frank Hahn for many hours spent in discarding false solutions, and of course to John Rawls whose profound work has caused us all to reconsider simple-minded utilitarianism. For further discussions of this and other aspects of Rawls's theory of justice, see Dasgupta [1974] and Arrow [1973]. Swed. I. of Economics 1973

324 K. J. Arrow and the validity of Rawls's reasoning from the original position to the maximin criterionl (see Arrow [1973] for remarks on some of these matters). One important aspect of justice is the intertemporal. What is a just allocation of goods among individuals of different generations? The intertemporal problem of justice differs from the contemporary problem fundamentally only in one empirical point: namely, that resources are productive, so that a transfer from an earlier to a later generation means, in general, that the later generation receives more (measured in commodity units) than the earlier generation gave up. In this case, our egalitarian presuppositions are somewhat upset; clearly, if we have any regard at all for the future generations (as justice demands) and if the gain from waiting is sufficiently great, then we will want to sacrifice some for the benefit of future individuals even if they are, to begin with, somewhat better off than we are. We will not do this indefinitely; this is usually formalized by assuming that they and we have diminishing marginal utility, so that at some point the gain in commodity terms ceases to be a gain in utility terms. We will introduce a very simple model of production to illustrate this point, and then analyze the implications of Rawls's criterion of justice in this context. We assume there is but one good. Within each generation, all individuals are alike, so that we may assume that there is only one. This also implies a stationary population; the adjustment to growing populations is easy and merely distracts attention from our analysis of the meaning of justice. The one good can either be consumed or used as capital which bears a return. Let Kt be the accumulated capital at the beginning of time period t. At that moment, part of the stock, ct, is consumed, and the remainder, Kt - ct, is used in production. Each unit so used yields a units at the beginning of the next time period, so that,
Kt+ -= (Kt -ct) (1.1)

Unless otherwise specified, it will be understood that the economy is in fact productive, so that one unit invested yields more than one unit of output, i.e., >1 (1.2)

The individual at time t derives satisfaction from the consumption, ct; it is assumed that the amount of satisfaction or felicity from any given consumption is the same for all t, i.e., there is a function U(c) such that the analogue of the sum-of-utilities criterion is the maximization of
t-O

Ut(ct)

1 Vickrey [1945, p. 329; 1960, pp. 523 f.] and Harsanyi [1953, 1955] have used an "original position" argument to justify maximizing the sum of utilities. Swed. I. of Economics 1973

Rawls's principle of just saving 325 where the variables ct are subject to the constraint (1.1) and the obvious conditions, Kt > 0, all t; Ko given (1.3)

However, it is reasonable to require that the future looks the same from any initial time-point. This implies that the sum-of-felicities maximand must have the special form,
t-O

2,l

U(cI)

(1.4)

for some fixed fi and some function U(c). fi is the subjectivediscount rate. The value U(ct) might be interpreted as the immediate satisfaction derived from consumption ct. A straightforward transposition of the Rawls maximin criterion to the intertemporal context might seem to be to maximize, min U(ct) (1.5)

subject to the constraints (1.1) and (1.3). This interpretation has been used by Solow [1974] though he recognizes that it is not that advocated by Rawls. It is pretty obvious (and will follow from some results below) that the maximization of (1.5) would lead to zero savings in every generation for there is no way to compensate the first generation for any saving they may do, and they would be worse off than any of their successors. Rawls's own discussion is contained in [1971, section 44]. It is indeed by no means clear. However, the one way in which he meets the point just raised is to argue that, "since it is assumed that a generation cares for its immediate descendants, as fathers, say, care for their sons, a just savings principle, or more accurately, certain limits on such principles, would be acknowledged" (p. 288). I will take up this sentence as the basis for a Rawlsian analysis of intertemporal justice.' It appears to me that in fact the assumption made uniquely defines a just savings principle, not merely "limits on such principles", and this savings principle so defined has much the same problems with it as the zero savings associated with the maximization of mini U(ct). Specifically, while savings may under some circumstances take place, it will always be followed by dissavings. Following the sentence just quoted, it will be assumed that the utility derived by an individual of generation t is a function, W(ct,ct+,) of its con1 John Rawls has informed me that he did not intend to supply any form of the maximum principle to intergenerational justice. But it appears to me that the logic which derives the maximum principle from the "original position" is equally applicable to intergenerational comparisons. Swed. 1. of Economics 1973

326 K. J. Arrow sumption and that of the succeeding generation. Then the Rawls criterion becomes, maximize min W(ct,ct+l) t (1.6)

For simplicity, it will be assumed, in agreement with Dasgupta [1974, assumption A.2.2] that W is additively separable in its two arguments, that the felicity ascribed by individual t to individual t + 1 is the same as that ascribed by individual t +1 to himself, that this felicity function is the same for all t, and finally that the felicity of the future generation may be discounted in the utility of the present generation. In symbols, W(c, ct+l) = U(ct)+fU(ct+,) As usual, it is assumed that, U(c) is increasing, differentiable, and concave (1.8) (1.7)

Finally, it is not excluded that i = 1 in (1.7); each generation might care for its heirs as much as it does for itself. We will, however, exclude the possibility, 2. Feasibility 2. Feasibility Conditions Conditions

We recall some simple restatements of the feasibility conditions (1.1) and (1.3). First, for any t, u, u > t, we can express capital at time u in terms of capital at time t and consumptions at times t, t +1,..., u-1:
u-1

Ku

u-t Kt-

v=t

C -CV

(2.1)

This holds trivially for u =t, and the general case follows from (1.1) by an easy induction on u, for,
Ku+1 = aKu - au = Ku - oU+l-toct-u cu;

if we substitute for Ku from (2.1), we deduce (2.1) with u replaced by u + 1. If in (2.1) we replace t by 0 and u by t, we see immediately that the condition (1.3), that capital never be negative, can be stated,
t-I
vO0

2 a-"cv< Ko,

all t

But since each term of the left-hand sum is non-negative, this requirement in turn implies and is implied by the familiar condition,
00

< 2 a-Vcv Ko v-O Swed. I. of Economics 1973

(2.2)

Rawls's principle of just saving 327 There is one consumption level, c, such that capital will be maintained intact, i.e., there is no saving. From (1.1), with t=O and K1=K0, we see that Ko= c(KO- ), or, C=(x )?K (2.3)

Clearly, at time period 1, since K1 =Ko, the consumption of c will cause K2 = K =Ko, and so forth, so that the constant consumption, c, will indeed cause Kt to remain constant at the initial level Ko. Starting from any given program, consider any alternative, say {ct} such that,
u-1
O t-vc=

u-1 at-V v=t

(2.4)

v=t

K. Variations satisfying then from (2.1), if Kt=K\, it must be true that K,=K leave the opportunities beginning at time u undisturbed. If, in particular, (2.4) u=t+2, we can say, if Kt= K and ct + o-t+ll =ct + a-lct+ then Kt+2=K;+2 (2.5)

We will refer to variations satisfying (2.5) as balancedvariations. By the same argument, an unbalanced variation will cause an increase or decrease in the capital stock. In the latter case, if it is repeated indefinitely, the feasibility condition (1.3) that capital stock be non-negative may ultimately be violated. In the following lemma, we consider variations from the constant-capital level of consumption. Lemma 1. Suppose a>l, and define c by (2.3). If, for some 6>0,
+ C2t a-lc2t+l > c(l + c-1)+6

for all t, then the program {ct} is infeasible. Proof. For any program, replace u by 2t +2 and t by 2t in (2.1).
Kat+2 = 2[K2t - (c2t + a-lC2t+l)] (2.6)

In particular, if ct = cfor all t, (2.6) holds with K2t+2=K2t=Ko. + Ko = C2[K- C(1 a-)] Let
cGt+ C-Cc2t+ - c(l + a-l) = ht

(2.7)

(2.8)
Swed. I. of Economics 1973

328 K. J. Arrow Subtract (2.7) from (2.6) and use definition (2.8).
K2t+2- K = x2[(K2t K) - ht] (2.9)

Define, Ko) xt = C-2t(K2tMultiply through in (2.9) by

Xt+ = Xto-2tht

CC-2t2.

Since
Xt=-

=0,
t-1 u=O

a -2Uhu

Suppose that,
t-1

lim sup
t--oo

c-2uhu >0
u=O

(2.10)

< Then, for some e >0, xt< -e for arbitrarily large values of t. Since oC-2tKO for all t sufficiently large, xt < - a-2tKo for some t. But from the definition of xt, this implies that K2t <0, and therefore the program {ctj must be infeasible. Under the hypothesis of the Lemma, h,u> for all u, and therefore (2.10) certainly holds. Corollaryto Lemma 1. The maximum feasible constant level of consumption is c, as defined by (2.3). Proof. If c > c, then c(1 + a-') > c(1 + a-); hence, by Lemma 1, the constant consumption level c would be infeasible. Lemma 1 can be generalized easily to cover consumption patterns for time intervals greater in length than 2. Lemma 2. Suppose > 1, and define c by (2.3). If, for some >0,
m(t+l)-I

2 v=mt

amt

m(t+l)-1

cv

2 v=mt

omt-v +

for all t, then the program {ct} is infeasible. Proof. Correspondingto (2.6), we note that,
[ m((t+l)- l

Km(t +l)=

Am

m(t
v-mtr

a mtvc

The rest of the argument is completely parallel.

Swed. l. of Economics 1973

Rawls's principle of just saving 329 Just for formal completeness, we examine the case where the economy is not productive in goods, i.e., a< 1. Lemma 3. If a < 1, then for any feasible consumptionprogram {ct},
t=0

: ct converges and therefore ct 0 as t- + oo.

00

Proof. For feasibility, we must have Kt+l > 0, and therefore, from (1.3), that Kt > ct. Hence,
Kt+l = (Kt - ct) Kt - ct

and therefore,
t-1

Kt < Ko-

cu
u=O

The condition, Kt > O, all t, then implies the Lemma.

3. Just Savings in a Utility-Productive

Economy

We shall say that an economy is utility-productiveif, given any feasible consumption program in which consumption is the same for two successive periods, t and t +1, individual t would prefer a balanced variation which increases ct+1and decreases ct. In other words, the economy is sufficiently productive, a sufficiently bigger than 1, that the individual would prefer to shift consumption to the next generation even though he may discount the felicity gains to some extent. Lemma 4. An economyis utility-productiveif and only if oa > 1. Proof. Start with a program where the consumption at times t and t +1 are both c. Suppose that individual t prefers ct, ct+,, a balanced variation of c, c, with ct+ >c. I.e., U(ct) +fU(ct+) > U(c) +fU(c)
Ct+ a-lt+l = c(1 + o-l)1

(3.1)

or (ct--c) + o-l(ct+1-c) = 0 (3.2)

Since U(ct)+U(ct+l) is concave in its two variables, from (1.8), (3.1) can hold only if the derivative of W at (c, c) in the direction of (ct, ct+i) is positive:
Swed. I. of Economics 1973

330 K. J. Arrow
U'(c) (Ct-c) +pU'(c) (ct+? -c) > 0

or, dividing through by U'(c) > 0. (ct- c) +f(ct+i -c) > 0 (3.3)

But (3.2) and (3.3) hold simultaneously, with ct+l-c>O, if and only if aoc> 1. We now note that for any productive economy, whether or not utilityproductive, the just savings policy must necessarily forbid any extended growth in welfare. The just savings policy will not necessarily require a constant utility level as defined by (1.7), but any generation which has a higher utility level must be preceded and followed by generations whose utility level is the minimum. Since more precise results will be obtained later, we give here only an informal sketch of a proof, though one that could be made rigorous. Theorem 1. Suppose a>l . Then if {ct} is a program which maximizes mint W(ct,ct+l), as defined by (1.7), over all feasible programs, it must be that (a) W(co,c) =mint W(ct,ct+), and (b) if (Wcu,cu+)>mint W(ct,ct+l), then W(cv, = v+1) mint W(ct, ct+l) for v = u- 1 and v = + 1. Proof. (a) Reduce co slightly and increase ct for t >0 by the income from the additional capital invested. If W(co,cl) were greater than the minimum, the latter would be increased if we make the reduction sufficiently small so that W(co,cl) does not become the new minimum. (b) Suppose W(c,, c,u+) and W(c,U+, c,+2) are both above the minimum. As in (a), reduce c,u+ slightly and increase all subsequent ct's. This will increase, and W(cu+,, W(ct,ct+l) for all t > u+ 1. The change will reduce both W(cu,cu+1) but for a sufficiently small change, they will still be above the original cu+2) minimum. Now increase ct for t < u by a small amount. This will reduce K,u+ proportionately. If c,+1 is then reduced by an equal amount, K+2, will be unchanged; the subsequent ct's remain feasible. Clearly, W(ct,ct+1)is increased for all t<u. Again, W(cu,c,,+) may be and W(c,U+, c,+2) is reduced, but by the change sufficiently small, they still remain above the original making minimum. Hence, W(ct,ct+l) is above the original minimum for every t, which contradicts the assumption that the original program was a just savings rule in the sense of Rawls. With the general flavor of the solutions now displayed, we give an exact characterization of the optimal policy. Theorem 2. Suppose oB > 1 and f < 1. Then the feasible consumptionprogram which maximizes mint W(ct,ct+l) can be characterized follows: Chooseco, cl, as to the constraint co+-c1 =c(l +ao-l); then at the to maximize W(co,cl) subject optimum, ct=c for even t, =cl for odd t. For this policy, the followingproperties [hold: c <<cc; W(ct,ct+)=mint W(ct,ct+l) for t even; W(ct,ct+l) mint W(ct,ct+l) for t odd, with the equality holding if and only if f = 1.
Swed. I. of Economics 1973

Rawls's principle of just saving 331 Proof. From the definition of utility-productivity and Lemma 4, c <c*. By construction of the program, W(ct,ct+i) is the same for all even t. Now let t be odd; then ct=c* and ct+l =c*. We know that U(c?)> U(co). Mutiply both sides of this inequality by 1 -B >0 , and do some transposition. + U(cO) lBU(cl)< U(cv) + fU(c*) = W(c.,ct+l); the inequality is strict if < 1. The left-hand side, however, is the common value of W(ct,ct+l) for all even t. Hence, the minimum is indeed achieved at every even t. Now consider any program, {c;}, for which mint W(c, c;+l) >mintW(ct, ct+i). In particular then, it must be true that for some e > 0,
W(ct, ct+l) > W(ct,Ct+) + E,

for all even t

But by construction of the policy {ct} the right-hand side maximized W subject to the constraint ct +ac-lIt+l = (1 + -l). Let b be the minimum value of c + a-lc subject to the condition, W(Co, > W(C,CL) C1) )+ Then clearly, b > c + a-c1, and, by definition of a minimum, ct+ o1-'c'+l> b= (c+ a-lc*)+ 5=c(l + a-')+b , >0

If we replace t by 2t in this inequality, we see from Lemma 1 that the program {ct} is infeasible, so that the policy {c'} is optimal. It is at least questionable that the saw-tooth pattern corresponds to any intuitive idea of justice.

4. An Extension of Concern to Further Generations: Approach to Utilitarianism Suppose that an individual is concerned not merely with the subsequent generation but with generations still further into the future, say m after his own. Then we might wish to generalize (1.7) to read,
m

W,= W(ct ..., ct+m)=

2 f=-

U(ct+t) flB

(4.1)

and we seek to maximize min Wt subject to the same feasibility conditions (1.1) and (1.3). Corresponding to the assumption of utility-productivity, we will assume that, starting with the same consumption at times t + i and t +j (0 < i <j m), the economy is sufficiently productive that generation t will i< prefer to shift some consumption from t +i to t+j, all other consumptions being held constant. The last qualification requires that
Swed. I. of Economics 1973

332 K. J. Arrow
-(t + )Ct+t-_(t + Ot+j +

be constant, from (2.2) (with equality, which will obviously always hold if there is no satiation in consumption). Hence a small increase h in ct+j must be accompanied by a decrease of at-'h in ct+i. Since all other consumptions are constant, an increase in Wt is an increase in f iU(ct+i)+#jU(ct+j). If in the initial situation ct+ ct+j=c, then a shift to later consumption increases Wt by,

U'(c) ,B ( - '-Jh)+fj U'(c)h

to a first approximation, which is positive if and only if fajoc>fifoc. I shall assume this inequality holds for all i and j, 0 < i <j < m. It will also be assumed that fBi>fPi+, a generalization of the condition < 1. This represents a non-increasing concern with increasingly future genf erations. Theorem 3. Suppose the utility of generationt is given by (4.1), that a&fiincreases with i for i < m, and thatfi is non-increasingin i. Then the feasible consumption program which maximizes mint Wt can be characterizedas follows: Choose c*(i =0, ..., m) to maximize W(c, ..., cm)subjectto the constraint,
m
E C m M-iCt= EM-t

2otct=c i0O

oai=-

where c is the constant-capital level of consumption. Then at the optimum, = (0< For this policy, the following properties hold:c* < lc* m). C(m)t+=cO i t+ c(M+1) i C ( i i < +1; Wt=mintWt if t is divisible by m + 1; for all othert, Wt> mint Wt, and the inequality is strict if ,fi >,i+ for some i <m. Proof. From the condition that aif is increasing, it follows as in Lemma 4 that c* <cci+1 <m). By construction, Wt is the same for all t divisible by m + 1. (i To show that the minimum is attained for all such t, it suffices to show that Wr> Wo(r m). The proposed program for generations r,...,m +r is, in order, c*, ...,c*,
co, ..., c*_. Then,
m-r Wr= E i U(?Cr*+) + m U(c_ m+rl)- =1 m-r r-1 i U(Cr*+i) +

i0m i-O r-1

i-m-r+l m

i=O r-1 fi0

itO m-r

2 ] ~+m-r+l

U(C

).

Wo= I ~t u(C?:)= :/,

m-r WrW W,= 1=0 ( i- tir)

i=O

(*) + 2 , u *)= )i=r r-1 U(C?+r)+ 2 (fi+m-r+1i=0

f,i U(C +

i-O

/, +r U(CT+r).

fA) U*(c,).

Swed. J. of Economics 1973

Rawls's principle ofjust saving 333 Since fl, is monotone decreasing in i and c* increasing in i, we see that and therefore (f,-fir)U(cir)>U(c*) ( -f). fj-fi+rt>0 Similarly, ~ i,+m-tr+l -fi 0, U(c*) < U(cr_x)fori < r - 1, andtherefore, (i+m-r+ )U(c) > -Bf). Further the first inequality is strict if O<i- m-r U(c*_-))(it+m-r+1 and fli >fi+r, the second if 0 i < r- 1 and fli >fli+,m-r+ Since fl non-increasing, we see that if fl, >Bj+ for any j < m, then fli >k for any i <j and any k > j + 1. By considering all possibilities, we see that if fj> f+l, some j<m, then one of the inequalities must hold. Wr- Wo AU(C*)+ BU(c*l) where,
m-r r-1

A-

2 (pi-i-0

+r),

2 B= (Bi+m-r+1-=i)
i=0

and the inequality is strict if flj>fj+ for some j< m. Also, A >0, and

A B=2 =2 i=0

m-r

-2

i=r

i-m-r+l

= o-2 -2

=
i=0

m i=0

=0

i=0

so that Wr- Wo> A[U(c*) - U(c*_)] > 0, with the strict inequality holding if the fli's are not all equal. Thus, the proposed program attains mint Wt at t=0 and all other t-values divisible by m +1. If it were not optimal, then there would be an alternative program {ct} defining utilities Wtwhich would have the property, among others, that Wt > Wo+ Efor all t divisible by m + 1, for some e > 0. The rest of the proof exactly parallels that of Theorem 2, the infeasibility of the program {c'} being established by Lemma 2. Thus, introducing altruism to generations up to some horizon leads to maximization of a sum of discounted utilities from the origin up to that horizon, though beyond that it leads to periodic repetition of the solution with a period equal to that of the horizon. Remark 1. If the horizon is taken to be infinite or at least equal to the horizon for the human race as a whole, then the maximin criterion becomes simply the utilitarian criterion, maximize Z flt=t U(ct). Under altruism toward future generations forever, the difference between the maximin criterion and the utilitarian disappears. Remark 2. The assumption that aifl is increasing over the horizon might be too strong if we take the view that the horizon is rather short. Since this product is zero beyond the horizon, it might be supposed to get small toward the end. Then c* is no longer increasing, and the reasoning leading to the inequalities Wr> WObecomes invalid. Some of these constraints become binding, so that the nature of the solution must change. It is an open question what the maximin policy is under these circumstances.
Swed. I. of Economics 1973

334 K. J. Arrow Remark3. Theorem 3 is valid for m =0, the case of no concern for the future. In that case, the maximin policy is the constant consumption c, as is intuitively obvious, thus justifying a remark in Section 1.

5. Unproductive

Economies

For completeness, I also consider maximin policies in unproductive economies. Only the case m =1 (concern only for the next generation) is considered. The economy may be physically unproductive, a 1, or unproductive in the utility sense, a > 1 but a/ < 1. The case a < 1 has in effect already been covered by Lemma 3. For any feasible policy, it must be that ct approaches 0 and therefore W(ct,ct+l) approaches its smallest possible value, W(0, 0)=(1 +/B)U(0). In this case, the minimum of W(ct,ct+1)may not exist; one can always find a feasible program with ct >O everywhere, for example, ct=coet, for any < a and co=-Ko(a-e)/a. But inf W(ct,ct+l) is defined. Theorem 4. If a < 1, then inf W(ct,ct+) The case cxf 1 is more interesting. <
=

(1 +-) U(O)for all feasibleprograms.

Theorem 5. If aB < 1 and > 1, then the feasible consumption program which maximizes mint W(ct,ct+l) is the constant-capitallevel of consumption in every period. Proof. Suppose false. Then there exists a feasible program, {ct}, such that W(ct,ct+i) > W(c, ) +6', some 6' > 0

Since W is concave in its variables and aWlact = U'(ct), aWact+l =PU'(ct+i), we have, W(Ct, (ct (ct+ct+) < U'(d) -) -+P) +fU'() so that,
U'(c) [(ct ) +(ct+l )] > '

-+

W(c, c)

or (ct -c) +(ct+ -) > , 6 >0 (5.1)

where 6=6'/U'(c). We now prove by induction the following two statements for each t: (a) K2t< Ko (b) (c2t-c) + a-l(c2t+1-)
Swed. I. of Economics 1973

> .

Rawls's principle of just saving 335 Clearly, if (b) holds, then by Lemma 1, the program {ct} is infeasible, completing the proof. Also, (a) holds for t=0. We shall show that, for each t, (a) implies (b), and that if (a) and (b) hold for some t, then (a) holds for t + 1. (a) implies (b): Clearly (2.2) continues to hold if 0 is replaced by any starting time t, and, in particular, by 2t.
00

-(-2t)c,
u=2t

< K2t < Ko

a-"-2t) >0

(5.2) and sum for u> 2t. (5.3)

In (5.1), replace t by u, multiply by

00 00

a--(-2t)(C
u=2t

+ - C-)

2
u=2t

-(u--2t)(cu,+ Cl-)

But,
00 00 00

-(u-2t)(c+1 u=2t

)=

-(u-2t)(--

C)

u=2t + 1

2 u=2t

-(

-2t)(u-

C)-

2(2t

Substitute in (5.3).
00

(1 + ao) I Since,

X(U2t)(cu-

u =2t

_) >Xf(c2t -c)

(5.4)

o'-(U-2t)= C
u=2t u-O

a-=

Ko

it follows from (5.2) that the left-hand side of (5.4) is non-positive. Therefore, c2t< , and, from (5.1), with t replaced by 2t,
C2t+l > c

(5.5)

A By hypothesis, ocp 1 or B< oa-1.From (5.1) and (5.5), (b) holds. and (b) imply (a) for t + 1: From (b) and (2.8), ht >O. From (a) and (2.9), (a) (a) holds for t + 1, completing the induction. References
Arrow, K. J.: Some ordinalist-utilitarian notes on Rawls's Theory of Justice. Journal of Philosophy 70, 245-263, 1973. Dasgupta, P.: On some problems arising from Professor Rawls's Conception of Distributive Justice. To appear in Theory and Decision, 1974. Harsanyi, J.: Cardinal utility in welfare economics and the theory of risk-bearing. Journal of Political Economy 61, 434435, 1953. Rawls, J.: A theory of justice. Harvard Press, Cambridge, Mass., University 1971. Solow, R. M.: Intergenerational equity and exhaustible resources. To appear in Review of Economic Studies, 1974. Vickrey, W. S.: Measuring marginal utility by reactions to risk. Econometrica 13, 319-333, 1945. Vickrey, W. S.: Utility, strategy, and social decision rules. Quarterly Journal of Economics 74, 507-535, 1960. Swed. I. of Economics 1973