If'-' working paper department of economics el of Social Insurans Variable Retiremei Diamond massachusetts institute of technology 50 memorial drive Cambridge, mass. 02139 A Model of Social Insuran Variable Retireme A P. A, Diamond and J. A. Number 210 Department of Economics, M.I.T, ** Nuffield College, Oxford Octob # \ .^" C' :V-P \y ^ \ 1 . Introduction No-one knows what work he will be capable of in the future. Uncertainty about earning ability in the last years of life is particularly great. The burden of this risk to the individual is eased both by private insurance and by the tax and social insurance system. Complete relief from risk is not available, because neither private insurance nor public arrangements distinguish fully between low income by choice and low income by necessity. might defeat itself through moral hazard. Full insurance For this reason, the design of optimal social insurance is an interesting and difficult problem. In this paper, we consider a simple form of earning ability risk, and 2 study optimal insurance for a population of identical individuals." We suppose that everyone lives for the same length of time, and that at any date an individual has either full earning capability or none: there is no partial loss of earning ability. strikes randomly. Loss of ability The government is assumed unable to distinguish those who cannot work from those who merely choose not to. Individuals are assumed to maximize their expected utility, and to . > be willing, without concern for the truth, to claim inability to work when it suits them. Three models will be used, with one period, with two periods, and with continuous time. The one-period model allows us to develop simply the basic condition determining whether or not there is a moral hazard problem. For the case where there is a problem, we relate the size of the optimal insurance plan to characteristics of the utility function. With the two-period model, we demonstrate an additional aspect of the moral hazard problem, that taxation of alternative - 2 - commodities can increase expected utility. For our model, we find that, under plausible assumptions, an untaxed individual would try to save too much. Thus the optimal social insurance plan needs to be supplemented by an interest income tax. In the continuous-time model, we examine the optimal consumption path while working, and the optimal relationship between pension and date of retirement. Even in the absence of utility-discounting, and of a positive return to saving, we find that it is optimal for consumption to increase with age, and for retirement benefit to increase with retirement age. This leads us to suggest a way of incorporating these conclusions into the benefit structure of the U.S. Social Security system. 2. 3 The One-Period Model Either the individual is capable of work or he is not. probability that he can work. He knows the He is an expected-utility maximizer, his utility being specified by three utility functions u. (c) • utility of consumption c :- when working, UjCc) utility when able to work, but not working, u,(c) utility when unable to work. When working, he produces one unit of output. Assuming nonlinear taxation and one type of individual, the lack of continuous adjust- ment of labour supply is not critical. On the assumption that work is unpleasant in Che aggregate, we have U2(c) > u (c) for all c (1) - 3 We also assume that work plus consumption is preferable to no work: u,(l) U2(0) > (2) There are no private insurance markets. The government pools risks well enough to he able to plan to distribute consumption, conditional on work, in such way that the expected value of consumption equals a the expected value of output tor any individual. The government being unable to tell why an individual does not work, its policy is entirely described by person who is working, and c«, c., consumption for a consumption for one who is not. This policy provides the individual with expected utility 6u (c.) 6u_(c^) - (I + (1 if he works 9)u,(Cp) - 0)u-(c^J if when he can he does not work Thus he is willing to work if and only if Uj(Cj) - u^ic^) It is as well to suppose individual works if (3) that, even in the case of indifference, he is willing and able to do so. the We shall comment on this assumption below. When individuals who can, work, the aggregate resource constraint is ecj since the output of a + (1 wcrker is - 0)C2 - 1. (4) e Otherwise, c_ = 0. The situation can be portr.ived in a diagram, with Cj on the axes. This is done in Figures 1 c and and 2, where the plane - 4 - is divided into Lwo regions by the curve assumption (2), the point = c u = u^(c„). By lies below this curve. c„ = 1, (c.) Everywhere on and below the curve, the individual works if he can; The feasible region is the shaded and above it, he does not work. Three indifference curves, area plus the origin- Ii^i* ^?^9» are drawn, along which expected utility is constant given the I_I., private decision on whether to work. curve = u^ u. It will be noticed that the lies entirely below the forty-five degree line: this expresses assumption (1). Evidently there are two cases to consider, which are shown in the two figures. In the first, the line with equation indifference curve 0c + (I ihe - indifference curve 9)c„ = 6. is tangent to Since the slope of the is "l^^l^ 6 1 and the slope of the budget (5) - 3 u^Cc^) line is - 6), - 9/(1 the optimum in this case is given by uj(c*) = u^(c*) Gc* + and (I - 0)c* =9 (6) In this case we have the familiar description of a full optimum. The moral hazard constraint is ineffective, as can be seen from the diagram: in the full optimum, If at we have Figure u 2 people are willing to work if they can. the allocation described by < u-,, (6) the situation is different, and must be as shown in Here the optimum is given by the intersection of the budget constraint and the curve u = u : F,louri F, 1 Z - 5 - Uj(c*) = u^fc*) For this case to occur, and 6c* + (1 - 6)0* = (7) 6 the indifference curve at this intersection This will point must slope down less steeply than the budget line. be the case if i u' on the contrary, case applies. > u! all points of the curve at u' at all u' = u„. u If, points of the curve, the first For convenience, ue state all this as a formal theorem. Theorem for all If 1 and x Uj(x) = u^(y) the optimum is given by = Uj(x) y implies u'(x) - u'(y) (7). for all If implies Uj(y) the optimum is given by x (8) and y u'(x) - u'(y) (9) (6). The conditions in the theorem take on a more interesting shape if we make the further assumption that the loss in the ability to work has an additive effect on utility: u^(r) = u^(c) - b, b - (10) Since in this case marginal utility of consumption is the same in the two states, u (x) = condition u„(y) It seems to us implies that than its contrary. additively separ.jblc (8) can be stated as u'(:) - u'(y), for all x and (II) y this condition will much more often be relevant For in example, it follows when utility is consumption and labour, and labour is disliked rioLire 3 r - 6 - 3 . Comparative StJtirs under By means of iht- dirtgrain, al Mt ^ we can i Ilaza'rd xaiTiine the v/ay in which the optii :inium changes when probabilities change, when the disutility of labour is changed, and when the utility of consumption is changed. When increases, the budget line becomes steeper, rotating 6 clockwise about the fixed point present, and c so that is c. u„ - u i.e. increased and c , , if moral hazard Thus, is Similarly an increase in the both incr-jase. c^ disutility of work, (1,0). moves the curve = u u downwards, decreased. A change in the utility of '.cnsumption can also be analysed by considering how affects the curve it u For example, we can = u„. show that an increase in risk-aversion can lead to a reduction in the extent of insurance, which may bt contrary to some people's intuition. Suppose utility and marginal utility at remain fixed, both with c and without work, and that risk-aversion increases, decreased for Therefore the and in particular for c^, ^ r. cur'.'e = u. u„ c optimal is decreased: claimed. utility at 4. thereby (See Figure 3.) . c is increased and the the extent of insurance is reduced, as A diiteitnt kind of ir^-rease in risk-aversion can have the opposite effect 3 is shifted to the right in the neigh- Consequently the optimal bourhood of c i? c u. the curve For example, by preserving utility and marginal while increasing, risk-aversion. ( u, = u The Two-Period is By analogy to Figure shiftec; up. .'lodel With two periods, we assume that everyone can work in the first - 7 - period, and that an itidividual is able to work in period two with probability fdcts are known to the government. rh'3se want to use nhis model to Wu slial 1 discuss the significance of private saving behaviour, but ior the present, ihe government is assumed to offer individuals pairs of first and second period consumptions: if work is done in period two, every case a The budget constraint c being assumed tVTO J - e)c (i t is a The = I + 0R (12) I aggregate output unity, and that there now in is iserved for derivatives. 1 ih^it periods. when everyone works is constant interest factor siiuation on ) taken to be is R[ec * . i Utility if not. ) .c be denoted by superscript instead funciii-'n will subscripts being of subscript, to portray ,c function oi consumption in the particular utility it ic (c R. In order budget constraint two-dimensional diagram, we use the when everyone works who is able {^2) to eliminate one of the variables'. It is most convenient to use : the a he variables where z o - c, = c, z and c (13) is the return to working, 1 - z the tax on labour income. c -^ = 2 Since consur.ipt u.n levels have so e * ^— R - 9(i - I c. - - (1 (U) e)z c t.' - (15) z) be nonnegative, long as everycne wciks when le can, and c it follows that, z are constraind o by three ILneai ; ni:qua 1 1 t ies , t - 0, c - 0, and c„ - 0. These define triangle in a However, people ,z)-space. (c willing to work when they can only if art; > 1 u ,c (c - ) u 2 (c If - z 0, - c c has disutility, implies that increases 2 u , . 1 , (c^,c^) ^o''2^ Thus the inequality (t^,c,). -0-1 u " > ' and decreases c, (16) ) (16) Notice also that, since an increase in 0- '> z z so that, maintaining our assumption that work , . ,c o 1 c-. satisfies (16) and (16) o (c ,z) if they can is the one bounded by z' satisfies ,z') (c l I > z v/henever The region in which people work z. ABC in Figure 4. If is ,z) (c not in this region, eilher there is no feasible allocation with these values of and c In the latter case, Gu^Cc ,c ) + - (1 o'^2' V = m z, or c ^ o e)u-^(c no-one works in the second period. Rr^ ,c,^). Max [9u^ * = and expected utility is 1, !.et - (I e)u-^ Rc^ + c : o = (17) i] / be the maximum utility obtainable without work in the second period The indifference curves drjwn in the diagran. are the curves of constant expected utility work 1 6u * the full optimum - ibe point attainable utility is F greater same argument as in Section unattainable ,x O where u^ 3 on the assumption that is not v is m , so feasible; and the maximum that moral hazard is The done in the second period. optimum sKows chat the full is if 1 (x 2 - th.^n present in the optimum, and worV 9)u The case sho\vn is one for which done in the second period is u - (1 ) I - u 2 (X o ,N,) _ < 1 ir.ilics u,(x 10 ,x, ) - 1 is the partial derivative with respect to 3 u ,x^) (x I O x.. i. (l: c /'': <;):('<:;• 4- -z [ I Cii^^i c S" - - 9 - We concentrate on the case wiiore moral hazard there work in the second perir-d at the optimum. is optimum the indifference curve small increase in z Then at the tangent to the curve is AB. An increase in reduce utility. v;culd present and is Also a may c o either increase or decrease utility, depending on the sign of the We shall discuss conditions for slope of the curves at the optimum. this in a moment The interesting point is that increase expected utility while change in a is z that would c held constant represents a savings opportunity that the individual would wish to avail himself of if he were allowed change in happen brings about c if it as can be seen from for, to; the i hanges in and c and (14) a (15), that would c^ were an act of pri\Mtc saving or dissaving. Thus in the opliKium cannot be obtained without preventing access to general, market, or, alternatively, the capita] imposing an appropriate tax or subsidy on wealth. It is most likely that the hazard constraint will have 2 If u u 12 the slope of the curve representing the moral negative slope at the point of tangency. a c tends to zero, then the curve AB only where = c = z 0. Since z c > is . at the point predominantly negative. is also zero, i.e. B, We can also rather plausible local cndiiion for the optimum to appear as in Figure U. For differ only by Theorem AB . can hit the line on the vertical axis, with a urve tends to minus infinity as = u find i 2 a this rurther analysis constant, Assume that assume that 2 u i.e., have the same derivatives. 3 u v.'o ^ = u' - b. and 3 u 10 - - 1 u (i) (x O ,x,) = u 2 (x I O < 2 u,(x ,x,) - u„(x ,x ] implies ,x„) / o 1 o I I ) I Then at the optimum, individuals want to save more, Our claim is that at the optimum a reduction Proof increase expected utility, curve u 12 = u m ..12 . is an increasing function of 3 9,1 (u dc c would being held constant; i.e. that the z has a negative slope at the optimum. increasing function of c Since we have to show that it z, u - u is an . - u 2,) I = u - u o 21.1 - ^(u K o o I - u 2. ) 2 Here numerical subscripts denote differentiation with respect to the indicated consumption level. We shall argue that, at the optimum, 2 1 u 1 = e_£ . (1 u - e)-|- (19) " Given- this equation, u than 2 — cc u ? 3 -^^ 1 - u") (u is positive if —r is larger 1 o (and conversely). u The latter conditions is precisely (ii) "2 It remains to derive optimum we have From Figure 4 a (19) given the as,-,umption that at the moral hazard problem (which follows from (i)). we know that at the moral hazard constraint. D an indifference curve is tangent to Equating these two slopes we have - 11 IN oRl 1 o,' - 6(1 o^. /, 12- 2 ,1 oR^ 11, e}(u[ - u^) oR2 - e)uj (1 12- ,2 oRl + &u^ Crossmult iplying, we obtain (19), which completes the proof. The first condition in the theorem is the one that we have earlier claimed to be plausible, and which is pretty much required The second condition for moral hazard to be present in the optimum. says that, at equal utilities whether working or not, the individual Our conclusion is has a greater incentive to save u'hen not working. that it is normal for a tax on saving to be associated with the optimal social insurance policy. Suppose now that saving is means that optimisation is ,il lowed, and is not taxed. further constrained by a This saving equilibrium condition, that the derivative uf expected utility with respect to is c Thus the condition is represented by the locus of zero. o points where the indifference curves have vertical slope, the curve in Figure 5. FE As one moves right along this curve, utility, and insurancL the size of the social diminish, programjT.e, If individuals made their savings plans under the assumption that they would work in the future, analysis xvould be completed by combining 12 the u savings constraint with the moral hazard constraint ,c,) (c o = u (c o i ,C-,) z and the 'iptimum would be at E, where the two constraints intersect. However if the individual plans savings and future work at the same tim.e, Max [eu'(c - s,c, o g - Max [eu^(c + s) v-. ^ have a differeni condition: (1 - e)u-^(c - s,c„ o 1 - s,c„ ' s) + (1 - e)u^(c * s)] 2 - s,c„ + s)] - This is 12 - when savings are optimal assuming work, a more stringent cond it ifir since the ability to adjust savings increases the utility available to those planning not to work.. is shown in Figure 5 as FG with the optimum occurring at model is explored in more detail in another paper.) when individuals are free to H. This (This Notice that, without constraint or taxation, it save- Q is still optimal to be on the boundary of the feasible set. One aspect of these solutioiis should be further noted. they are on the boundary of the feasible set, fact indifferent ^'hether to work in period individuals are in Similarly in or not. 2 Since the one-period model they are indifferent between working and not working. If this indifference w., re reflected in random choice rather than an implausible adherence to government wishes, there would, though the government could adopt strictly speaking, be no optimum, policies arbitrarily close to thf ones we have discussed under which individuals would certainly wish to work it is not inappropriate to assunn , if That is why they could. as we have done, ment can choose when the consumri' is indifferent. that the govern- Of course a fully satisfactory treatment of these matters would have to model the costs of implementing choices explicitly, and that would take us fo fara field. The chief conclusion of this section is that a government concerned about social insurance would normally want to discourage private saving, for example by taxing it; and that, do so, the extent of the social diminished. This exemplifies situations, that the provider a o] if it could not insurance programme would be general feature of moral hazard insurance would usually want to control trade in related commodities. For example, fire insurance - - 13 companies should want their clients to buy fire extinguishers; and they like to see high taxes on 5. t'lie consumption of tobacco. The Continuous-Time Model We turn to model in which the a etirement date i is continuously variable, while maintaining the assumption that labour choice at each This model moment is disctetc. to ask two further .illows us questions: how consumption shouM be made to vary with age when working, and how the retirement benefit should depend upon the age Since the moral hazard constraints can no longer be of retirement. pictured in a diagram, the mathenat ics of the model is more complicated than for the previous models, but we shall use the earlier analysis to guide us to the solution. We make the follov;ing assumptions. output per unit period would be unity. zero. Utility taken to be is taneous utility. Ivhen the tlie everyone were working, If The real interest rate is undiscountcd integral of instan- indi'.idual is working, his utility is according u.(c), when not working, his utility is u (c ) u^(c) or All three functions are strictly concave. to whether he is able to work cr not. /To simplify the analysis, ve assume from the outset that u„ (c) Saving is = u. (c) controlled by and u., b, + differ by a a Consequently, function of age, c consumption when retired can be a function both of age age at retirement, becom.es unable to r, c^Ct.r'). work is a The date constant: b - 0. government. th^i consumption when working can be u- s (t); t at which an and and of individual random variable, with density function f - and distribution function and f(s) for '^ < < - The length of life is denoted by F. s 14 T, There are no atoms in the distribu- T. tion. If s, the age of retirement is r and the age of disability is utility is r T s /uj (C|(t))dt /u^{c,(t,r))dt + o " r - s T r = /u^(C2(t,r))dt + /U|(t)dt + /u,(t,r)dt - b(T - s), " o (20) r where we have used (19), and introduced the notation U|(t) = U|(Cj(t)), U2(t,r) = U2(c2(t,r)) Neither individual nor government can affect the last part of (20)- What we are interested in is or r v(r) = /uj(t)dt + T /u2(t,r)dt (21) and expected utility can be taken to be r V(Cj,c,,r) = ,*" /v(s)f(s)ds v(r)[] + - F(r)] (22) o since the individual has to retire at the date of disability he has not already done so. In '22), individual decides to retire if he is r is s it che age ac which the still able to work at that date. (22) v, expresses the maximand for our problem as an average of the expected value of frequently m v(Min I's.r)). our analysis, and it is Tliis average occurs convenient to have e special , , 15 - For any integrable function notation for it- /h(s)f(s)ds = (h) J + h(r)[l - h, we define F(r)] (23) For later reference we note the following Properties of (i) J (ii) J is a positive J h is a (iv) h If is is and (i) - we let r' -^ dif f erentiabl (iv) J,(h)-J^(h)= '^ Thus , (h) ^ r where h and only if (h) = h'(r)[l - F(r)] (24) follow immediately from (23). / [h(s) - h(r)]f(s)ds To prove (ii) that [h(r') - h(r)] /f (s)ds + (25) r' (h) J h if h is monotonically increasing. The case r is monotonically decreasing From (25) constant. a constant, it also follows that To complete the proof of for all dif f erentiable, and constant. if r '^ J r e and calculate from (23) r, r constant. dr^r Proofs r. monotonically increasing (or decreasing). constant function of is a (h) linear functional for each mono ton ica 1 ly increasing (or decreasing) in is (h) if (iii) J r. (24) is J proved similarly. (h) (iii), is constant if suppose that From (23) it follows that then implies h' = 0, i.e. h(r) h J h (h) is is a is = c, 16 - - The net resource cost of an individual who retires at becomes unable to work at s is T r z(r) = /cj(t)dt requires a /c2(t,r)dt + o recalling that and r - r (26) r worker produces unit output. If the government units of output for its own purposes, the aggregate A resource constraint is J (z) + A - (27) We are We can now state formally the problem to be addressed. to Maximize J subject to (27) and (v) J|.(v) - J^(v), all t - r (28) This last constraint says that individuals do not prefer to retire before r. Since the government can set consumption levels equal to zero for anyone who retires after a date it chooses, we need not be concerned about individual plans to work too long. effective constraint on government is that it choose functions c. and In the full, c„ so that (27) and (28) Thus the r, and the hold. first-best optimum, the government is constrained only by (27), and marginal utilities of consumption should be equated in all circumstances; c,(t)=c°. If C2(t,r) - c°; under these circumstances we had with u.(c ) u| (c°) - - u^ (c") (29) u„(c„), there would - 17 - be no moral hazard problem, for, as one can see from (21), would be nondecreasing function of a and r, J To see this consider the derivatives of J consequently (v) nondecreasing function of t. there is no moral hazard problem. In the case where / the optimal value of J Using property (iv) of r. ^^(v) = ^^(2) Thus TT—J with u,(c dr r II ) = u„ (c L i. c ) = {c° - c° - and ; , {u,(c°) - U2(c°)}{l unless > implies = ul u; (v) J - c^, r = T, » ^—J or r 1}{1 normally with (z) we have - F(r)} - F(r)} except in the borderline case r (z) normally, e.g. when - A - or if a r r respect to is r and (v) v(r) 0, for r - A Cannot Thus it (c° - c°-l){/sfds + r(l - F(r)]}. i. J o ' in these cases be optimal to have r < T, since an increase in would increase utility without increasing the resource cost. = J r r (z) = c°T + z \ The interesting case is where u - u„. u! = u' inconsistent with is Accordingly we assume from now on that U|(Cj) = \i^{<i^ implies uj(cj) -U2(c2) Before proceeding with the analysis, we note feature of the optimum. Once a man has retired, a (30) convenient there is no advantage from inefficient intertemporal allocation of his consumption. Incentives to work depend on expected utility as a function of the retirement date, so the cost of providing expected utility conditional on retirement should be minimized. C2(t,r) = C2(r) Thus (31) T. 18 - This means that and v z can be written r /U|(t)dt v(r) = + U2(r)(T - r) (32) o r z(r) = /cj(t)dt + c^CrXT - r) - r (33) o The bad feature of the maximization problem we have to deal with is a lack of the concavity conditions that would render necessary conditions for the optimum also sufficient. The moral hazard constraints (28) cannot be expected to be concave in any of the control variables u- c., c„, as the control variables, or But if we regard r. the maximand, and the constraints (28), are linear, and the resource constraint is convex in them. (27) Thus we have a well-behaved programming problem if given. and u r is taken as We shall first derive sufficient conditions for the optimum conditional upon a specified value of and afterwards derive a r; necessary condition for optimality with respect to r. The earlier models yielded solutions in which the individual was indifferent whether to work or not. expect that in the present model, about the age of retirement. the individual will be indifferent We first consider the best path with this property and then show that it is is a constant independent of independent of r, t, then by property (iii). /uj(t)dt Correspondingly, one can + indeed optimal. v(r) is a If J (v) constant We write U2(r)(T - r) = V (3A) o Given that v is constant, we ask what is the optimal combinati on . - of and u. maximizes Let It is the one that for u„. the given value of or equivalent ly for given v, and G. G be : /Gj(u,)dt + v minimizes he inverse functions of C|(t) ^^GjCujCt)) u^Cc), so that 2(r) « 19 - - o /ujdt])(T and u.(c) - r) (z) (z) c^Ct) =G2(u2(t)). and G^ij-^iv J J Then - r (35) o and it follows that 'Gj(U|(t)) - GjCu^Cr)) Consequently for 5u (t) ^r^^^ " (t - r) (t > r) t < r ^fSl^'^^ " g2(s)]f(s)ds + [g,(t) - g2(r)][l - F(r)], (37) where we have introduced the notations gj(t) - G|(U|(t)) = I/u|(c,(t)) (38) g^Ct) - G'Cu^Ct)) = l/u^(c2(t)) For cost-minimization, the derivative (37) should vanish for all t; r / (g,(t) - g2(s)]f(s)ds + [gj(t) - g2(r)][I and using a • Differentiating with respect to (1 - F(t)]^,(t) o - F(r)] (39) for a time derivative c, we obtain [g^(t) - g^(c)]f(t) which can be written equivalently as = 0. (40) \ )r \ Xq, = a / V/ \ A . // / ^ f \Ciur e 6 / .-'-'U,='^2 n -/ / " -^ - 20 i'-f<"i^;ry<r7 t)) Also, setting = t in r " "[uj(cj(t)) u^^c^Cr.))]^^''^ (39), we have g,(r) = g2(r) (Al) or, equivalently, u|(c,(r)) = u^Cc^Cr)) We have shown that if it is true that there is indifference about retirement age at the optimum, then and (Al) constant, and (AO) is We shall show that this is indeed hold. to do this we need I,t v to know a little more about satisfies two differential equations, differentiating (3A) with respect (T - t)u2(t) to the optimum, and the proposed solution, (40), and one we obtain by r: = U2(t:) - Uj(t) (42) In Figure 6, we have a phase diagram for this pair of differential equations, in ,c„)-space. (c functions of c and c^. functions of c. and c^ and u and g g„ u„ are of course increasing are also increasing respectively; since is equal to g i/u;(c,). From (40), c_ u. c. is is stationary when = u„ stationary when u ~ ^y lies below the curve the two curves, both and c. g. g = g,; By assumption = g^ (i.e. = r. Therefore both throughout the solution. c and the curve (30), u! = c„ g Between u'). are increasing with c„ says that the solution we propose has to hit the t while from (42) = t. g„ (41) curve at are increasing functions of t . - 21 - Theorem 3 Assume that implies = u„ u T - dt '^ ^^1 g, r. ' ^2^1 = gj at - F r the policies so defined are optimal for J (z) + A = 0, given t f th.. and If u^ - u, dUj dt - u'. u! If the optimum for a given Assuming that u (0) = u-(0) stated properties exists, for any = -«, r exists, it is unique. r a solution with the and between T such that r / o (i.e. (1 - F(s)]ds > A (43) such that positive consumption is feasible). We first proved this theorem by introducing Lagrange multipliers for the constraints, and following the standard method of proving sufficiency of the first-order conditions for constrained maximization problems. here, This method is rather involved. The proof we give though less clearly motivated in that Lagrange multipliers are not introduced explicitly, is some preliminary results. The policies satisfying the conditions of briefer. the theorem are denoted by asterisks. We begin by establishing The outcome is compared with alternative policies satisfying the constraints of the maximization problem. * Lemma J (g^v) > - g * (0)J (v) for any v satisfying J (v) < - J (v) . 22 - * Since increasing function of is an c t, is an increasing g 1 function of and we have t, /g,(t)[J^(v) - J^(v)]dt - (4A) o Now ^* /gjJ^(v)dt = - [gj(r) g,(0)]J^(v) (45) and ^ * r ^ fk]\iv)dt= r ^ t /g,(t) /v(s)f(s)dsdt O /gj(t)v(t)[l - F(t)]dt + o ' o T = r /[g*(r) - g*(t)]v(t)f(t)dt /[g*(t) + o . - g*(t)]v(t)f(t)dt o by reversing the order of integration in the first integral, and using (40) in the second integral. This last expression simplifies to /g*Jj.dt = g*(r)/vfdt - since Lemma = g^fr) Combining ("44), gj (r) 2 Jj-Cg*) (45), /g2vfdt and (46) = g*(r)J^(v) - J^(g*v) we obtain Lemma (^6) 1. = 8*^°) The second and third conditions of the theorem are equivalent to the vanishing of (37) for all result. t. Putting t = 0, we obtain the stated - 23 - Proof of the Theorem We now show that any feasible path does not have higher expected utility. From the definition of we have z /[Gj(Uj) - Gj(u*)]dt z(s) - z*(s) = + [G^Cu^Cs)) - G2(u*(s))](T - s) o Since G and G_ are convex functions of their arguments, we can use the inequality for convex functions to obtain g z(s) - z*(s) - /g*(t)(Uj(t) - u*(t))dt + g*(s)(u2(s) - u*(s))(T - s) o Using the definition of v, v (s), we (32), and the constancy of can write this as g z(s) - z*(s) - /(g*(t) - g2(s))(uj(t) - u*(t))dt o * T / ££Jll(u o 9U|(t) = (t) + g2(s)(v(s) - V*) - u*(t))dt + (A7) g*(s)(v(s) - V*) ' The second step follows from the expression for the derivative, We now wish to apply the linear operator (36). J first term on the right hand side vanishes since (37) all (47). is t T "" to o ^ au,(t) Thus we have T ' ' o / it \ ''\9u*(t)/ ' ' The zero for . - 24 J/g2)v J^(z) - J^(z*) - J^(g*v) - - g*(0)[J^(v) by Lemmas 1 and 2 J (z) + (A8) A < Now - v] A - 0, and J (z ) + A Therefore (48) implies = 0. that J - V, (v) . r expected utility i.e., is I no greater on the alternative path than on the one satisfying the conditions of the theorem. This proves the sufficiency part of the theorem. We next prove the existence part of the theorem. Refer to the diagram. Since marginal utilities go to infinity as consumption goes to zero, g and g then go to zero, and the g = g. curve Similarly the curve u = u passes passes through the origin. through the origin. The value of g,(r) determines the point on the curve at which the solution path ends, and therefore choosing the whole path. small enough, aggregate consumption, g,(t) be made as small as we please, uniformly for J (z(t) + t) = (t) choosing + r - rF(r) /[I - F(t)]dt = z(t) + t, can - r. Thus Therefore g,(r) if (43) holds, J (z) (z) + + A can be made negative by small enough. J r g,(r) on integrating by O ^ We now have to show that choosing t By r /tf(t)dt O parts. - = g„ can be made as small as we please. r J g large enough. A can be made positive by By equation (40), we have - 25 - g - - since to g,/g~ ~ r, we and -^ log 1 - F [1 - F(t)], decreases with Integrating from t. obtain g,(0) - g,(r)[l - F(r)] Therefore by choosing - r, - J without limit. (z) large enough, c (0) , and all c (t) can be made as large as we wish, thus increasing for t g,(r) (49) The proof of existence is complete. The rather restrictive condition about utility functions introduced to prove existence is by no means necessary. As will be evident from the proof, the condition was introduced merely to exclude cases with zero consumption, and their attendant corner conditions. Throughout the argument, r has been taken to be less than T. The form of the differential equations makes it clear that the case r «» T would need careful handling; and the above existence proof would not go through at all. most plausible cases, Fortunately, it turns out that, in the the optimal value of r is less than T. This will emerge as a by-product of our discussion of the optimization of r. We have shown that, for a given value of r < T, an optimum exists. u,'(c (r)) = and for a given The optimum is given by a solution of our differential equations which at time satisfying A, u'(c (r)). r takes values c,(r), c„(r) Denote utility arising from the 26 - - (40)-J42) the solution to / c J by (z) Z(r,x). formal proof, that and = x (r) V(r,x), and denote the value of by We first note, without taking the space for and V are dif f erentiable functions of Z X. for given If, Theorem 3 Z(r,x) + A = 0, it follows from satisfies x r, that the solution path so defined yields an optimum. utility is strictly concave, the optimum for given Therefore the equation Z(r,x) A + Z(r,x(r)) for the set of Lemma 3 ' x(r) r is a (50) is unique. r has a unique solution, and = A = + (50) which are feasible. continuous function for implies that x(r) is T. r < To prove this, we must establish that when r - r < T, If x(r) by the equation we can define a function by r is r restricted From the bounded. differential equations we have the inequalities ^1 dt Integrating from g,(0) > to 1 - F' T - dt t we deduce that r, g,(r)[l - F(r)], U2(0) > U2(r)(T - r) (51) - g,(r)[I Also - F(;)] - are related by lower bounds on all It follows c that equation = u! (t), 'r) c^(0) - c^Ct) - c^Cr); and C|(0) - c'j(t) - Cj(r), c.(r) = X U2(r)(T - C2(r) Therefore we have upper and u'. which are increasing functions of (50) consider a convergent sequence ^ imposes an upper bound on r -^ V and r < o r(< T) . Since Now x. r V - r x. - 27 - eventually, Z(r ,x being ° continuous, we have and equal to x(r and has a limit point, is bounded, {x(r )} o o ) continuous function of r, and that, as is a dif f erentiable function of derivative. 3, x o V(r) = V(r,x(r)) is unique r is a varies continuously, r We next establish that all vary continuously. c.(t) Thus 0. Z . This proves the lemma. ). The lemma implies further that optimal A = + x V(r) and obtain a formula (66) for its The argument used is similar lo that employed in Theorem where we expressed z(s) the functions in terms of and u v. In the present case, we must identify dependence on the retirement date r, We have and this will be done by superfixes. s /c[(t)dt = z'^(s) + C2(s)(T - s) - (52) s o with c^(t) = G|(Uj(t)) :2(s) = Gi7T-7lV(r) \ /u[(t)dt}j. - o Since the resource constraint is fixed, \,(z^') for any r, r' < and eventually let - J^iz"") = (53) We shall manipulate the left hand side of T. r' Since the function ->• (53) r. u is so chosen as J^(z^) - J^(z) to minimize J (z ), (54) 28 - where /c^(c)dL = :(s) + Gj~--{V(r) /u[ (t)dt})(T - - s) - By the mean value theorem, we have z(s) - z'^'cs) = where T g„ J V o to lies between (r,r',s) as (s) G'f^r^{V^(r,r',s) (55), / u| V(r) and ' (t)dt }j{V (r) - V(r')} V(r'). - r'; and r and use r r g^(s) to (5A) (53), - {J (z'^') r r as r' -* Apply the operator r. : (g^) + £,(r,r')}{V(r) z - V(r')} (56) 1 e(r,r')-*Oasr'-*r. where Reversing the roles of (g„) is continuous J^Az"") - where £2(^»^') -^ in Jj.Cz'') as and r r', and using the fact i\(^&l) *e2(T,r')]{V{v) - V(r')} - r' that we also have r, - (57) r. Applying the mean value theorem to the left hand side of (56), recalling property (iv) of J that , (42) = = c[(s) - C2(s) H^(s). = h'(r)[l - F(r)], and and using the tact, derived frc,n/(52), ^"(s) (h) ir—J ' ' + that g2(s)[u2(s) - U|(s)] - 1 (58) say, l-F(r) we obtain J^,(z''') - J (^'^^ = H'''(r)(r' - r). (55) here tends to G' / J .(z'"') - J J - (59) - 29 - with between r and r By the continuity of consumption r'. levels both with respect to time and with respect to the terminal date, we see that H r ' r ^ (f) as H (r) - r' imply that (59) {H'(r) + n,(r,r')}(r' - r) - where both and e. r H (r) and and G- (60) and - r) , (^^.(82) J/g^) (g "i to r' V,r') - V(r) < Letting 0. =• ) * e, {V(r') - V(r))/(r' - r) is equal -* H r (r)/J as r (g,). ' '' r' r, r' < -^ r ) {V(r) - V(r')} (60) . also obrain from (60), r. ' {V(r ) } ) - V(r')} Now consider (62) tends to r from below. is dif f erentiable, r' > (61) r. Recollect that g (62) ^ ^2 implies that the limit ot r from above exists, and i ' '- r tends to the same limit that and that -LilL (r), defined H "2 We have therefore established V'(r) = - We now express * A similar argument wiih as tends to "'('^^ jy,) {V(r') - V(r)]/(r' - r) way. ' * £2 ^r ,r establishes that r' ) ,,r ' t J r V "'^"^^ since ' , imply (61) ,r - (r ,r e r, we tend to zero as r\ + r', and using the continuity of and r with respect to (g„) + n2(r,r')}(r' {H'^(r) where r J {\(gl) tend to zero as n, Reversing the roles of V(r) Therefore (56) and r. + (r) = g^(r). by (58), (63) in a slightly different Then we can write - 30 - = {c^(r) H^'Cr) - = {- h,(r) g^(r)u[(r) + h^Cr) - - c^Cr) + - F(r)] l}[l F(r)] - 1}(1 - g''^(v)u^^{r) (64) where we define h.(r) = gj(r)uj(r) - c^(r), r = 1,2 u.(c^) 1 r 1 - c. evaluated ac r (65) 1] (66) u.(c.) Thus we have, finally, the formula V(r) = ' "/^^^ h ,(r) - We must now consider the sign of g = g , h,(r) h - + h^ + along the curve I since that curve contains the locus of end-points for r-optimal paths as r A straightforward calculation yields varies. dh. -— From this it follows that, as ^(h, - u. - h^) u < u„ on the curve The inequality (68) (c ,c ) on the curve which therefore (c (r),c (r)) = u, to = - U2 (68) 0, g . implies that there is ac most one point g V'(r) = lies g varies, = g^ g = g, < since (67) the = g at which if c^(r) = c., h left of that point - h, i = + = I 1,2. V'(r) 0, and at If > 0; if it lies . - 31 V (r) to the right < * * exists for which T for If 4 exists a (r . = i ul(0) = «, - a when Consider first how Since defines r u (°°) I = 0; and if there -^ = u' (69) u|(c,) = u^(c2) 1, behaves as h.(c.) 11 Let 0. u! ^ hj(Cj) - h2(c2) = - number. this 1,2, such that c^ c,, 1 g = such that > then there exists equivalently, i , . 1,2, u^ ~ u Proof. * = c. ) We have not argued that there is at most one value satisfying C (r) = c r Lemma c - * r . < r the optimal policy. of there exists such a point, and if there If 0. c. -» (70) 0, or, 1 be an arbitrarily small positive e is concave,' u. 1 + u. (c.) u.' (c.) (e - c.) - u. (e) Therefore u.(c.) > -> ^(O - e as c -> . 1 It follows that if u. is 1 while if u.(0) - 0, it 1 negative for small is also Lim h.(c.) 11 c., * 1 Lim h true that Lim [hj(Cj) - h (c )] 11 . (c = . ) = 0. = 0; Thus (71) g-^o Now consider what happens as since Lim ul = 0, g -^ <». c , Fron (68), c -+ (69), °°; or, equivalently, and the assumption that - 32 - g -> - h h °°, decreases without limit: Lim [hjCcj) - - = h^ic^)] ~ (72) g-«o since ~ h„ h continuous decreasing function of is a hold, a solution of (72), Lemma (70) must exist, and tlie g, and (71) lemma is proved. Let 5 _ (T - t)f(t) A^^^ <l>(t) _ F(t) , . ^^^) , be bounded as t -^ T. x(r) If Z(r,x(r)) A and T / o < feasible), [1 - F(t)]dt Let A some M defined, by A = 0, (so that positive consumption is then x(r) Proof + is as °o ^- r - T (7A) be a positive number, and consider paths optimal for with c,(r) - M On any optimum path, number a, g dependent on > g , u„; and there exists a positive < u. but independent of M < of these paths and for all Max {g| If this were not so, have g - g and (75) t, - g^, - t r, such that for all < - r, Ut - Uj} - a (76) then by compactness and continuity, we should for some r and t u„ - u / which is impossible by assumption. From (76) and the differential equations satisfied by the - 33 - optimal path, follows that it + u^) ^(g, = - (g, $ "i^r^ ^) ^ a Min (-pi-^, Since ^"2 " &2^Y^* (77) defined in (73) is bounded, there exists (T - t)f (t) - F(t) 1 and therefore (77) d < ^^ , V > 2' a N f - F 1 .-l^log Integrating from to t Therefore if r g^(t) as -> -F(t)] ~ log [1 - F(r)] - ^ log F(t)] - [1 and the corresponding r-optimal paths all -> T •• u^(t) -< - ^o - T, and r [1 we obtain r g,(t) + U2(t) - g|(r) + U2(r) + c^(t), C2(t) such that 1 implies dt'^1 satisfy (75), N > Thus t < T. for any fixed T - F(t)]dt. J^(z) -> - / [ It i o follows that for all sufficiently large r, inconsistent with (75). ->• x(r) "» a weak one, automatically satisfied if zero limit as result of Lemma T. -* t 5 + (z) A = is as asserted. should be noted that the assumption that It rather Therefore J (p f(t) is bounded is tends to a non- would have to be very pathological for the f to fail. When the assumptions of these Lemmas are satisfied, it follows that the optimal r is less than T. For when x(r) > c, , . . - 3A - V (r) 0, and this must be < V Since is value of true for all sufficiently large evidently continuous at T T, r. cannot be the optimal r Summarizing the above analysis, we have Theorem Under the assumptions of Theorem 4 3, assumption (69), and the assumption that marginal utilities tend to zero as consumption levels tend to infinity, there exists an optimal The optimal path or paths is Identified by the unique (c-,C2) satisfying the additional condition < / T. r h^Cr) - h,(r) = where h are defined h„ , It should be recognized that that optimal less than is r quite plausible- It T in (78) 1 (65). the condition is states that the marginal disutility of labour bounded away from zero for large consumption. T is to infinity as t used to prove excessively strong, though is the optimal (69) r, it will be optimal to have tends to T. In cases where c and tend c Such cases are therefore rather peculiar 6 . Features of Solutions Several aspects of the optimum emerge from the general analysis and call for explicit notice. In the first place, the consumer is indifferent about since v is constant, the date of his retirement. In effect, he is assumed to retire when the government wishes him to. A small deviation from the optimum would make the consumer strictly prefer r to any other retirement date. But the government wishes 35 - consumers to retire before the end of life (and could set consumpThis latter tion for later retirers to zero to achieve this). achievable is situation where the feature contrasts with the /full optimum'in which, normally, everyone who can work does. Surprisingly, that curious feature of the real world, the compulsory recirement age, may have some optimality properties. Secondly, since the lifetime utility of a man who loses his ability at s is is borne by the v(s) - (T - s)b, sufferer, the entire cost of ill-health but this is the only source of difference among the utilities of different individuals. If b were zero, everyone would have the same lifetime utility, while the marginal utility of income consumption in different periods would vary from individual to individual. Insurance would be perfect in a naive sense, not the economist. sense. 's Thirdly, it may seem curious that the solution is independent of b, that is, independent of the effect of ill-health. is that marginal The reason utilities are, by assumption, unaffected by the state of health of the consumer. difference between u and In the more general varies with u. c , case where the the manner of dependence does affect the optimum. Fourthly, it is worth summarizing the basic features of the optimum that emerge from the diagram: U2(C2) ^'' u,(c,) (79) Thus an individual would always be immediately better off if he retired. He is prepared to continue working because of improved retirement benefits with additional work. It may be helpful consider why one would not want Starting from such a u^ ~ "i* to - 36 - situation, an increase in decrease in at c retirement date, financed by a earlier, increases expected utility from planning c transfers consumption to the state retirement at that date, since it with higher marginal utility. Thus work is not discouraged, and The argument works at utility increased. r, and therefore also at earlier dates. We also have u^Cc^) > uj(C|) (80) Thus the individual would prefer more insurance. the extent of social Moral hazard keeps insurance down. We recall also the basic intertemporal facts, that c« and c are different iable functions of time, and that dc ^>0 (81) -7-^ ^ at (82) There is no difficulty in principle about calculating optimal policies. First the terminal point of the optimal path has to be found, by solving simultaneously u! = u' and h, - h = are calculated backwards from there with various values of each case, J (z) is a Paths . r. In calculated, and we have the solution for that particular value of the resource constraint. procedure with 1 simple case. We illustrate the - 37 - Example = log c, u. u„ = a + log c, =1, f(s) It and is readily checked that Theorems c., 3 1 and 4 ^2 t)— c - (I - t)C| The terminal point is given by Cjdog = c„ c. - Cj = 1) - = log — "l + In terms of apply. the differential equations in Theorem (1 so that T = a > 3 c are a c = Cj - C2 = c„ c. c^dog c^ and a - + + 1) = 0, I 1/a. To solve the differential equations, we define — ^2 *^1 w = log . , m = log (1 - t) - log - r) (1 In terms of these variables, we have dw dm -r— =e ""W -w+a-1, w = ^^^2 —— =c(w-a), dm z I c.=— 2 a m at at m = = Solution of these equations gives the optimal path for any value of r. For each r we then compute little manipulation, to be J (z), which is found, after a - 38 - J (z) = ; o r - (1 t)(c „ = (1 In this case, + c„ - it a -log(l-r) - r)^[ / then, is ^^ -^- l)dc + Z \ -„ "^ (c- + c-e 9„, - 1 De'^^dm + -] relatively easy- to obtain the solution for different resource constraints, since only one solution of the differential equations In table is required. we give solutions for I 12 a = .5, of the government net subsidy to the scheme, Table II displays solutions for of work, a = 1 , a and three values -A, .5, 0, case with much greater disutility and the same values of the net subsidy. be noted that in table I, c. -.2. and and are equal to c„ 2 It should at r. This is a very high figure relative to the wage, which is unity. rather implausible assumption about the disutility of work is required to reduce c (r) and c-(r) below unity. But in all cases people receive consumption during most of working life that is less than the wage plus the net subsidy, and only a very small proportion of the population obtain a retirement benefit that is greater than wage plus net subsidy. Two other features of the numerical results deserve comment. In all to cases c - c^ increases with t until t is very close r. The second feature is that in table when there is a I, very large government deficit r is nearly (A = -.5). 1, even Thus the direct effect of moral hazard, arising from the consumer's ability to retire healthy without special penalty, is almost A - 39 - Table a = -.5 .3 1.00 1.03 1.05 1.09 .4 1. 13 .5 .6 1 .7 I .1 .2 .24 .32 .9 .95 .99 .77 .78 .50 .81 .53 .54 .56 .59 .62 .66 .73 .86 1.01 1.02 1.45 1.69 1.92 .8 .5 1. 12 1.34 .64 1 .38 .39 .40 .42 .43 .45 .47 .51 .2 .23 .24 .24 .25 .26 .27 .28 .31 .32 .33 .34 .35 .37 .40 .44 .51 .969 = .30 .56 .66 .77 1.13 1.46 r = A A = .83 .86 .90 .95 1.17 I .31 .33 .39 .46 .67 .51 .60 .88 = .000 (more accurat .998 r - 1 ; 1 1 .8 X 10 Table 11 a = _ A_ = A_ = .2 ^1 .87 .1 .2 .3 .4 .90 .93 .96 .99 .59 .63 .69 .75 .85 .5 .6 '2 .49 .52 .54 .57 .7 .8 .91 .30 .33 .35 .37 .40 .44 .50 .59 .78 .96 .51 .65 .9 = .494 r == .17 .18 .19 .20 .31 .95 .99 r ^2 '^l .28 .30 .32 .34 .36 ,39 .44 .61 .66 -72 .80 ' .882 r ! .21 .23 .25 .29 .35 .49 .76 = .963 - 40 - entirely eliminated: hardly anyone retires early. outcome Nevertheless the far from the first-best optimum that would be achievable is if individuals could be prevented from retiring except through ill- health; for in that case consumption would be a constant independent of age and state of health. 7 . Private Saving As in the simpler models, we can ask whether individuals would attempt to alter their consumption plans tiiey if thought they could lend or borrow at the zero interest rate which the government Calculating the gain from savings, we shall show that faces. individuals would choose to save. A consumer working at date date t' wants to lend for repayment at t if t' u|(C|(t)) < [uJ(C|(t'))(l - F(t')) + /u^(c2(s))f(s)ds]/[l - F(t)] When the opposite inequality holds, the consumer would choose to lend for repayment at date Dividing by t' - t t'. and letting t' -* t, we find that the consumer wants to save if ^;(c,) >r^(u; - u-) (83) Only if there were equality here would the individual be content with the saving being done on his behalf by the government. Reference to Theorem 3 shows that the optimal policy does not eliminate incentives to save or dissave; for in the optimum. - 41 dt u 1 - - F u u Now a simple calculation shows that , ^"1 1, ,,2,1 ^"P ^Tr^-^) - — "P^ ^^ , *^i -^2 , (85) "2 i Applying this inequality to (84), we obtain ^; >i4t(u; -u') (86) This proves that the consumer wants to lend, that is, to save. The government can prevent this happening by suitable taxation of saving. It is interesting to enquire what the government should do if it has chosen, market. or is constrained, to allow a perfect untaxed capital Then in equilibrium ^u; -^(ul = - up (87) and it is also the case that uj «= u^ at r. (88) This last equality holds because of the possibility of saving for retirement, or borrowing against retirement benefits from the period when one knows one will be retired. We conjecture that the third-best optimum, when private saving is unconstrained, is defined by (87), the consumer retiring age. be (88) and the condition that indifferent about We have not been able to prove this. - 42 - The Magnitude of Transfers 8. An insurance scheme said to be actuarially fair if variations in is the date of retirement have no effect upon the net discounted transfers to the individual. The optimal social insurance scheme The net we have derived is not in this sense actuarially fair. discounted transfer to an individual retiring at notation, Theorem (see equation z(s) For 5 < s r, in our (33)). is z is, s decreasing function of a s when That is, those forced to retire social insurance is optimal- earlier receive larger net transfers from the government. Proof Since z(s) /C|(t)dt = C2(s)(T - + s) - s, o z' (s) = Cj(s) - C2(s) By Theorem 4, r, z' = of (89) 0, , the right hand side of and negative on the point. u^ 2 1 ' ^ + u' c^(s){T - The situation is shown in Figure (c ,c„) obtaining u.. 1, (89) I is zero at the optimal 7, where the curve labelled at which the right hand side vanishes, is shown sloping downwards and to the right from differentiate the right hand side of = - (89) the terminal point of the optimal path. u - curve to the left of the terminal = u' which is the locus of s) - u u - c„ = c + - (u„ - u )ul,V(u') 2 . To justify this, we can (89) with respect to c„, This is positive above the curve Consequently points between the curves u' = u' and , - A3 for which = u u is c less than (r) c all have z'(s) < 0. Since the optimal path lies entii'ely in this region, the theorem is proved. The meaning of this result is that those who are unfortunate enough to suffer disability early in life receive a larger net transfer from the State than those able to work until late in life. The optimal social insurance scheme subsidises those who retire early, though only to the extent compatible with maintaining incentives to work. The following result is of interest. Corollary to Theorem 5 Under the optimal policy, for all C| (t) i.e. - c^Ct) < t < - r, I the extra consumption obtained by working is always less than the marginal productivity of labour. Proof By the Theorem, We have also shown that c 9. - c„ - 1 < 0, the right hand side of u„ > u (89) is on the optimal path. negative. Hence as was claimed. Social Security in the United States Despite the level of mathematical complexity, the models we have considered are very special. It would be silly to base policy on the particular equations we have derived. the current U.S. There are two aspects of publicly provided pensions which it may not be premature to criticize on the basis of the analysis we have performed, c, F' I C/if rt 7 - 44 - First we shall put the models briefly in perspective. If individuals are saving rationally, the incentive for work comes from the change in their lifetime budget constraints with additional work. they If are consuming their net wage, it is necessary to consider separately the incentives which come from current wages for work and from increased future pensions as result of additional work. a they If are following savings rules which differ from both of these models, the two parts of compensation matter differently, although not necessarily in the manner we have analysed. The U.S. population no doubt contains individuals whose behaviour is describable by a wide variety of models, and not just the fully rational model. In the model analysed we found that a growing pension benefit permitted a higher pension, relative to the wage, than would be possible otherwise given the moral hazard problem. However if a pension which grows with work done is to serve as an incentive for work, individuals need to be awaie of the relationship between pension and work done- l^ile it would be expensive, a greater flow of information from the Social Security administration to individuals Hearing retirement age about their own pension prospects might well be worth the cost - Under the current U.S. system individuals receive their pension independent of whether they continue working once they reach age 72. Before that age there is an earnings limitation on pension eligibility. In terms of our notation the pa^inent of benefits independent of retirement represents increase in c., a large and discontinuous the consumption enjoyed while working. analysis we found that a growing level of the growth should be continuous, c c In our was optimal but that can be increased continuously - 45 - either by reducing taxes repeatedly for those continuing to work or by paying a steadily growing fraction of pension benefits independent of any retirement test. 15% of benefits might be paid For example, ac age 66 independent of retirement and 85% subject to the retirement The former fraction would grow steadily to test. 100% as in individual ages to 72. 10. Conclusions The main results we have obtained may be summarised by highlighting what they would recommend if they were applicable to the real world. Many of the results suggest practicable policy changes, and may be worthy of study in more realistic models. (1) The presence of moral hazard implies the desirability of policies that leave consumers indifferent about the date of This conclusion might be modified in models where .different individuals begin with different abilities, or' probability retirement. distributions for the loss of earnings. (2) It is an essential adjunct that capital to an opcimal insurance scheme transactions by consumers be taxed. cases, this would be a In most positive tax, discouraging saving. This conclusion might be modified if we allowed for nonrational saving behaviour. (3) Post-retirement benefits show a strong tendency to increase with the age of retirement. (4) Contributions co the insurance scheme should diminish with age, sometimes quite rapidly. eventually. This is They may become negative not as odd as it seems, for the optimal - 46 - allocation can be accomplished alternatively by paying part of benefits independent of retirement. The Corollary to Theorem shows that the additional consumption from working rather than retiring immediately should never exceed the marginal product. (5) There are some indications that contributions diminish more rapidly, and benefits increase more rapidly, as age increases. (6) Under an optimal scheme, the incidence of retirement for reasons other than ill-health may be very low, even when the utility gain from healthy retirement is quite large. this, the form of the optimal Despite scheme may be very different from a first-best redistribution between people of different ages that ignores all incentive considerations. (7) An actuarially fair scheme may discourage retirement less than an optimal scheme. 5 - A7 - Footnotes Financial support by the N.S.F, during the preparation of this 1. paper is gratefully acknowledged. Computing assistance was provided by the Computing and Research Support Unit of the Oxford University Social Studies Faculty. In another paper, we consider a diverse population, whose 2. savings opportunities are limited only by linear taxation. The theoretical effects of social insurance on retirement have 3. been studied by 4. Feldstein (1 974) and Sheshinski (1 977) . The model under consideration is equivalent to the standard maximization of an additive social welfare function. The particular distribution of characteristics assumed here has not, to our knowledge, been investigated previously. * 5. For example, by preserving utility and marginal utility at while increasing risk aversion. u = u„ is By analogy to Figure 3, c the curve shifted up. equals the inverse of one plus the interest rate. 6. R 7. In addition it will be necessary to restate the moral hazard constraint to allow for the combined choice of savings and labour plans. 8. To achieve the optimum at D in Figure 4, the government must either close the capital market, or introduce nonlinear wealth taxation to ensure that individuals do not desire to save. , , - 48 - 9. The analysis of optimal r lengthy and rather technical. is The impatient reader can turn to Theorem at the end of this 4 * section, The length and delicacy of the which states the essential results. argument arises pa'rtly from the difficulty of establishing that maximum utility is a dif f erentiable function of r - if it were not, unpleasant possibilities would arise - and partly from the difficulty of discussing the case r = T directly. feature is that, under plausible assumptions, condition for optimal r a The surprising simple first-order can be obtained, and defines optimal r uniquely. 10. We would not expect this result without the assumption that the marginal utility of consumption of a nonworker is independent of his health. I 11. To put it another way, higher retirement benefits are an incentive to work in every earlier period, the later the date the more periods in which work 12. If is encouraged. the utility function with labour were y log c + log (2 - y) implying unit labour supply in the absence of lump sum income, the disutility of work is log (2 - 0) - log (2 - I ) = log 2 = .6931 This Cobb-Douglas model probably overstates the utility of not working at all A9 - References Feldstein, M, 1974, , Social security, induced retirement and aggregate capital accumulation, Journal of Political Economy Sheshinski, E. , 1977, A model of social security and retirement decisions, mimeo. 3131jlY Z86102 i^^kH-. 1^ t,\!V( li'^ MIT LIBRARIES 3 3 3 TDflD DDM Mb 3bE mS TQflD DDM m5 TDflD DDM MH 3 4 375 3fl3 LIBRARIES MIS 3T1 TDflD DDM win LIBRARIES 3 TDflD DD H ms MO^ l/^r^ LIBRARIES 3 ^OflO 004 qirSYlp 3 TDflD DDM MIS MaS