ANALYTICAL OPTIMAL CONTROL THEORY AS APPLIED. TO STOCHASTIC AND NON-STOCHASTIC ECONOMICS 2 = ROBERT COX MERTON S., Columbia School of Engineering & Applied Science (1966) M.S., California Institute of Technology (1967) SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS’ FOR THE DEGREE OF DOCTOR QF PHILOSOPHY at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY September, 1970 ignature of Author ......22......2.. precise eer Department of Economics, August 28, 1970 ertified by ..... ener Thesis Supervisor mae ecepted by ......55 pert ei an wet Pen Chairman, Departmental Committee on Graduate Student's ~ ~Dewey GS. ST aap OCT 22 1970 Eieranies(44) ABSTRACT ANALYTICAL OPTIMAL CONTROL THEORY AS APPLIED TO STOCHASTIC AND NON-STOCHASTIC ECONOMICS Robert Merton Submitted to the Department of Economies on August 28, 1970 in partial fulfillment of the requirement for the degree of Doctor of Philosophy. The thesis consists of five self-contained essays that have in common that each is an economic's problem in intertemporal maximization. Chapter I examines the lifetime consumption-port- folio problem under uncertainty where the individual acts to maximize the expected value cf an integral of utility of con- sumption over time. It6's Lemma and the Fundamental Theorem of stochastic optimal control are introduced as the means for analyzing the problem. The advantages of working with the continuous-time version are discussed, and a number of theorems are proved. The emphasis of the paper is on efficient portfolios and the effects of uncertain time horizon and wage income. Non- stationary and non-Markov expectation mechanisms are introduced. Chapter II studies the same problem as I, but with different emphasis. The stochastic Bellman equation is derived in a heuristic but informative fashion. Similarly, the use of limits of discrete formulations eliminates the need for the formal use of stochastic differential equations. The explicit’ dynamics of the optimal rules over the lifetime of the individual are exam- ined in detail, The technique of comparative statics is used to study the effects of changes in risk, return, and risk-aversion on the consumption-saving decision. Asa by-product, generalized intertemporal income and substitution effects are defined. The infinite horizon problem is studied in detail. Chapter III considers the Ramsey-type social planning problem in a neoclassical growth model where the rate of population growth is endogeneous and is a function of per capita wealth or income. Under the Samuelson-Diamond criterion of maximizing social wel- fare, it 1s shown that the turnpike is no longer at the golden rule capital labor ratio. For the Bentham-Lerner criterion, it is shown that the origin of utility of the representative man is not arbitrary in the sense that the optimal program is not independent of this origin. It is further shown that the Schumpeter zero-interest level is never an optimal steady-state. Chapter IV is a study to find the equilibrium price re~ lationship between two perfectly correlated securities. The particular security examined here is a warrant. Based on ex- pected utility maximization, explicit formulas are derived for the warrant price as a function of the common stock price and the warrant's maturity date.In 1965, P.A. Samuelson presented a simple theory of warrant pricing. For a special class of warrants called per- petual warrants, the theory leads to an explicit solution for the warrant price. In chapter V, an econometric investigation of the theory is made. Because of the "tight" specification of the model, the emphasis was on statistical testing of the assumptions of the model and the model itself. The model was pitted against alternative theories in a forecasting "contest". ‘The results were quite favorable for so closely a specified model. Thesis supervisor: Paul A. Samuelson Title: . Institute Professor(iv) ACKNOWLEDGEMENTS I owe an irreducible debt to Paul A, Samuelson for intro- ducing me to mathematical economics in his course where I learned the basic techniques used throughout this thesis; for introducing me to economic research as his assistant. for two and one-half years which lead to my first published econ- omics paper, a joint paper with him; for the countless hours of his time which he made available for discussion of further research which lead to this thesis and two more published papers; for financial assistance through his National Science Foundation grant which allowed me the maximum time to pursue this research. I gratefully dedicate this thesis to him, I would like to thank Robert Solow and Franklin Fisher for helpful discussion and for serving on my thesis committee. Chapter V is based on work done for Fisher's Econometrics course. Peter Diamond read earlier drafts of Chapter I, David Scheffman was a faithful listener and helped in some of the mathematical proofs. Stanley Fischer provided useful suggestions for the work in Chapter II. I thank M.I.T. for financial support under a NDEA fellowship and the National Science Foundation for both a fellowship and aid under grant GS-1812. Chapter II was originally published in the August, 1969, Review of Economics and Statistics. Chapter IIT was originally published in the December, 1969, Western Economics Journal. Chapter IV is a shortened version of a paper with the same title, written Jointly with Paul A. Samuelson, and published in the Winter, 1969, Industrial Management Review. Sue Friedman and Jacquelyn Tricomi did an excellent job of typing a hard manuscript under rush conditions. I must thank my wife, June, for her patience and her support during the many evenings and weekends I spent doing this re- search over the last three years, and for s0 much morc. Without her, none of this would have happened.(v) TABLE OF CONTENTS Page No. Chapter I OPTIMUM CONSUMPTION AND PORTFOLIO RULES IN A CONTINUOUS-TIME MODEL .........eeeeeeeee 1 Chapter II LIFETIME PORTFOLIO SELECTION UNDER UNCERTAINTY: THE CONTINUOUS-"IME CASE .......... 48 Chapter III A GOLDEN GOLDEN-RULE FOR WELFARE-MAXIMIZATION IN AN ECONOMY WITH A VARYING POPULATION GROWTH RATE .. hee eeeeeeeeeeeeeeeeeees 73 Chapter IV A COMPLETE MODEL OF WARRANT PRICING THAT * MAXIMIZES UTILITY ....... beet ee eeees vee 88 Chapter V AN EMPIRICAL INVESTIGATION OF THE SAMUELSON RATIONAL WARRANT PRICING THEORY ........... 112Optimum Consumption and Portfolio Rules in a ContSnuous-time Model® Robert C. Merton M.I.T. March, 1970 1. Introduction. A common hypothesis about the behavior of (limited liability) asset prices in perfect mar- kets is the random walk of returns or (in its continuous- time form) the "geometric Brownian motion" hypothesis which implies that asset prices aré stationary and log-normally distributed. A number of investigators of the behavior of stock and commodity prices have questioned the accuracy of the hypothesis. ‘1 tn particular, Cootner [2] and others have criticized the independent increments assumption, and Osborne [2] has examined the assumption of stationariness. Mandelbroit [2] and Fama [2] argue that stock and commodity price changes follow a stable-Paretian distribution with infinite second-moments. The non-academic literature on the stock market is also filled with theories of stock price patterns and trading rules to "beat the market", rules often called "technical analysis" or "charting", and that presupposes a departure from random price changes. * T would like to thank P.A. Samuelson, R.M. Solow, P.A.Diamond, J.A. Mirrlees, J.S. Flemming, and D.T, Scheffman for their helpful discussions. Of course, all errors are mine. Aid from the National Science Foundation is gratefully acknowl- edged qQy For a number of interesting papers on the subject, see Cootner [2]. An excellent survey article is "Efficient Capital Markets:A Review of Theory and Empirical Work" a draft), E.F. Fama, November, 1969.In.an earlier paper [12], I exaitined the: continuous- time consumption-portfolio problem for an individual whose income is generated by capital gains on investments in assets with prices assumed to satisfy the "geometric Brownian motion" hypothesis. I.e. I studiéd Max EJ U(C,t)at where U is the instantaneous utility function; C 1s consump- tion; E is the expectation operator. Under the additional assumption of a constant relative or constant absolute risk- aversion utility. function, explicit solutions for the opti- mal consumption and’ portfolio rules were derived. The changes in these ‘optimal rules with respect to shifts in various parameters such as expected return, interest rates, and risk.were examined by the technique of comparative statics. : The present paper extends these results for more gen- eral utility functions, price behavior assul ions, and for income generated alse from non-cepital gains sources. It is shown that if the “geometric Brownian motion" ny~ pothesis is accepted, then a general "Separation" or "mutual ° fund" theorem can be proved such that, in this model, the classical Tobin mean-variance rules hold without the ob- Jectionable assumptions of quadratic utility or of normality of distributions for prices. Hence, when asset prices are generated by a geometric Brownian motion, one can work with the two-asset case without loss of generality. If the fur- ther assumption 1s made that the utility function of the individual 1s a member of the family of utility functions called the "HARA" family, explicit solutions for the opti- mal consumption and portfolio rules are derived and a num- ber of theorems proved. In the last parts of the paper, the effects on the consumption and portfolio rules of al- ternative asset price dynamics, in which changes are neither stationary nor independent, are examined along with the effects of introducing wage income, uncertainty of life expectancy, and the possibility of default on (formerly) “risk-free" assets. *-3+ 2. A digression on Tt6 Processes. To apply the dynamic programming technique in a continuous-time model, the state variable dynamics must, be expressible as Markov stochastic processes defined over time intervals of length h, no matter how small h is. Such processes are referred to as infinitely divisible in time. The two processes of this type '?’are: functions of Gauss-Wiener Brownian motions which are. continuous in the "space" variables and functions of Poisson processes which are discrete in the space variables. Because neither of these processes is differentiable in the usual sense, a more general type of differential equation must be developed to express the dynamics of such processes. A particular class of contin- uous-time Markov processes of the first type called 16 Processes are defined gs, the solution to the stochastic differential equation 3 qQ) aP = £(P,t)dt + g(P,t)dz where P, f, and g are n-vectors and z(t) is a n-vector of 2) - 2'1 Agnore those infinitely divisible processes with in- finite moments which include those members of the stable Paretian family other than the normal. 3) 318, Processes are a special case of a more general class of stochastic processes called Strong diffusion processes (see Kushner [9], p.22). (1) is a short-hand expression for the stochastic integral t t P(t) = Plo) + £ £(P,s)ds + f a(P,s)az where P(t) is the solution to (1) with probability one. A rigorous discussion of the meaning of a solution to equations like (1) is not presented here. Only those theorems needed for formal manipulation and solution of stochastic differential equations are in the text ani these without proof. For” a-complete discussion of 1t6 Processes, see the seminal paper of Ité [7], 1t6 and McKean [8], and McKean [11]. For a short aéscription and some proofs, see Kushner [9], p.12-18. For a heuristic discussion of continuous-time Markov processes in general, see Cox and Miller [3], chapter 5.standard normal random variables. Then dz(t) is called a. 4 multi-dimensional Wiener process (or Brownian motion). The fundamental tool for formal manipulation and solu- tion of stochastic processes of the It6 type is It6's Lemma stated as follows > : Lemma: Let F(P},.+.,Pyst) be a C? func- tion defined on R"X[o,@) and take the stochastic integrals - Py(t) = Py(o) + fF £,(P,sdas +f ey (P,s)az,, 2 = 2,...4n5 then the time-dependent random variable ¥ = F is a stochastic integral and its stochastic differential is = 9n BF ar longn __2?F ay 2 3, ap, + 55 dt + 22, Sn BPP, apap, where the product of the differentials aP;dP, are defined by the multiplication rule o @z,dzy = py at 4,J = 1,...,n on dz,dt = 0 ied, where p,; 1s the instantaneous corre- lation coefficient between the Wiener Processes dz, and dzy. yy 8 dz is often ferred to in the literature as "Gaussian White Noise There are some regularity conditions im- posed on the functions f and g. It is assumed through- out the paper that such conditions are satisfied. For the details, see [9] or [11]. (5) 9'see McKean [11] p.32-35 and p.44 for proofs of the Lemma in one- and n-dimensions. 6) 6 This multiplication rule has given rise to the formalism of writing the Wiener process differentials as dz, =7,vae where the S, are standard normal variates (e.g. sée [34).Armed with It6's Lemma, we are now able to formally differ- entiate most smooth functions of Brownian motions (and herice integrate stochastic differential equations of the 1t6 type). ‘7 Before proceeding to the discussion of asset price be- havior, another concept useful for working with It6 Processes is the differential generator (or weak infinitesmal operator) of the stochastic process P(t), Define the function G(P,t) by (2) c,t) = limit z [2 P(t+h),t+h)- P(t) t) ] h+.o nh when the limit exists and where " t is the conditional ex- pectation operator, conditional on knowing P(t). If the P,(t) are generated by It6 Processes, then the differential generator of P, Lp» is defined by eyn a 2, lynn a? GB LptUT ty ey t get 2k dy Ay 3P,3P; where £ © (frye), EF (GreeeeoGy)s arid ay; = 8,6;0y3- Further, it can be shown that gay Gee) =L,cacr,t)3 G-can be interpreted as the "average" or expected time rate of change of the function G(P,t) and as such is the natural generalization of ut ordinary time derivative for deter- ministic functions. (7) M warning: derivatives (and integrals) of functions of . Brownian motions are similar to, but different from, the rules for deterministic differentials and integrals. For example, if tard ik P(t) = Plo)e To4% 2b = p(oyeZ(t)-z(o)- Bt then aP = Pdz. Hence s§ $2 = stax # log (eceeto). Stratonovich [15] has develaped a symmetric definition of stochastic differential equations which formally follows the ordinary rules of differentiation:and inte- gration. However, this alternative to the Ité formalism will not be discussed here. 8) 5’, heuristic method for finding the differential generator is to take the conditional expectation of dG (found by It6's Lemma) and Mdivide" by dt. The result of this operation ° will beZ ,[G], ive. formally, L E,(a@) = @ =<,t61. The "{p" operator is often called a Dynkin operator and is written as "Dp",Asset price dynamics and the budget equation. Throughout the paper, it is assumed that all assets are of the limited liability type; that’ there exist continuously- trading, perfect markets with no transactions costs for all assets; that the prices per share, {P,(t)}. are generated by It6 Processes, i.e. ap. is Pot G) Fy a,(P,t)dt + o,(P,t)dz, where we define a, = f,/P, and oy 6,/P, and a, is the instantaneous conditional expected percentage change in price per unit time and o,? is the instantaneous conditional variance per unit time. In the particular case where the "geometric Brownian motion hypothesis is assumed to hold’ for asset prices, a, and 0, will be constants. For this case, prices will.be stationarily and log-normally distributed and it will be shown that this assumption about asset prices simplifies the continuous-time model in the same way that the assumption of normality of prices simplifies the static one-period portfolio model. To derive the correct budget equation, it is necessary to examine the discrete-time formulation of the model and then to take limits carefully to obtain the continuous-time form, Consider a period model with periods cf length h where all income is generated by capital gains and wealth, W(t) and Py(t) are known at the beginning of period t. Let the decision variables be indexed such that the indices are imple- coincide with the period in which the decisions mented. Namely, let N,(t) = number of shares of asset i purchased during period t, ivez-between ¢ and t + h (6) and C(t) = amount of consumption rer unit tine during period t The model assumes that the individual "comes into" period t with wealth invested in assets so that mM Wee) = AP ny Cony, ()Notice that it is Nj(t-h) because Ny(t-h) is the number of shares purchased for the portfolio in period (t-h) and it is P,(t) because P(t) 1s the current value of a share of the 1th asset. The amount of consumption for the period, C(t)h, and the new portfolio, N,(t), are simultaneously chosen, and if it is assumed that all trades are made at (known) current prices, then we have that (8) e(tyh = SIPEN, (t) - Ny(t-n)]P, (t) The "dice" are rolled and a new set of prices is determined, P,(t#h), and the value of portfolio is now J)” N,(t)P,(tth). So the individual "comes into" period (t+h) with wealth W(t+h) = xt N,(t)P,(t+h) and the process continues. Incrementing (7) and (8) by h to eliminate backward differences, we have that (9) -C(tth)h ya (N, (t+h)-N, (t) JP, (th) =D UNy Cot) Ny (t) IEP, (t4h)-Py (6)] + SP Ong (ttn), (£) 17 (6) and (0) W(t#n) = IP Cee 9 Taking the limits as h + o, we arrive at the continuous version of (9) and (10), ' = an. ca") ~C(t)dt = yt aN and C20") WCE) = SIP Ny (EPC) t) aP t) £) + WP any (t) ‘De use here the result that 1t6 Processes arc right- continuous (see [9], p.15) and hence P,(t) and W(t) are right-continuous. It is assumed that €(t) is a right- continuous function and throughout the paper, the choice of C(t) is restricted to this class of iUsing. It6's Lemma, we differentiate (10') to get =n yn n ql) aw ye NyaPy + Yh) aNgPy + PTT anyary n D ay The last two terms, Jj! aN,Py + 51° dN,dP,, are the net value of additions to wealth from sources other than capital gains. 1°’ Hence, if dy(t) = (possibly stochastic) instan- taneous flow of non-capital gains (wage) income, then we have that be pn n (a2) dy - c(t)dt = Jo aNyP, + Dy ajar, From (11) and (12), the budget or accumulation equation is written as Woe pn - , (3) au = YY Ny (tary + dy - C(t)at It is advantageous to eliminate N,(t) from (13) by defining anew variable, wy(t) = N,(t)P,(t)/W(t), the percentage of wealth invested in the ath asset at time t. Substituting for dP,/P, from (5), we can write (13) as = 37 Ts in qua) ai = SiiwyWaydt - c at + ay + ays! 93 mn. n 11 where, by definition, J)" wy zl Unts) vestion seven, Tf fll be aes See. a1) dnesne Js derives + sitke La owe of vie ueassees Is ‘risi-free" (by convention, ¢ n®2 asset), then 9, = 5 the instantaneous rate of return, + Will be called r, and (14) is re-written as 7 =u" a eae my GM) av = SYM wy (as-r uae + (eH-c)at + ay + ET p58 Tov A0"This result follows directly from the discrete-time ax gu- ment used to derive (9) where -C(t)dt is replaced by a genera] dv(t) where dv(t) is the instantaneous fiow of funds from ail capital gains sounees. Tt vas necessary to derive (12) by starting with th discrete-time formulation because is is not obvi the continuous jon directly whother dy-c(t)at equals m OGNLP sn Pe or 4 Lar DY ashy + SP anydPy or just yy aN,Py. nay There are ro other restrictions on the indiv because borrowin and short-selling arc allowed. omUsing It6's Lemma, we differentiate (10') to get =n n n qi) aw = SYP NSP + Dy aMyPg tL agar + yn n The last two terms, x aN,P, + xy aN,4P,, are the net value of additions to wealth from sources other than capital gains. ‘1°’ Hence, 1f ay(t) = (possibly stochastic) instan- taneous flow of non-capital gains (wage) income, then we have that (a2) ay ~ C(t)at = YP aNyPy + YyP aN,aP, From (11) and (12), the budget or accumulation equation is written as (3) aw = FP Ny (tary + ay - C(t at It is advantageous to eliminate N,(t) from (13) by defining a new variable, w,(t) = Na CODPSCEI/MCED the percentage of wealth invested in the 1£f asset at time t. Substituting for dP,/P, from (5), we can write (13) as = TW - n, Zz (ayy aW = SViw,Waydt - C at + ay + Siwy Woydz, = 7,9)" ae t 311 be assumed that dy where, by definition, J)” Until sectton seven, tt os d.e. all income fs derived fren canitel rains on aes Y ir one of the n-assets is "risk-free" (by convention, the ntB asset), then a, = 0, the instantaneous rate of return, ,, will be called r, and (14) is re-written as nl =v" m (uit) dW = Sy wyCay-r)Wdt + (rW-c)at + dy + SP) Myoqday 30. i0\onis result follows directly from the discrete-time argu- ment used to derive (9") where -C(t)dt is replaced by a general dv(t) where dv(t) is the instantaneous flow of funds from all non-capital gains sources. It was necessary to derive (12) by starting with the discrete-time formulation because it is not obvious from the continuous, version directly whether dy-C(t)dt equals A n hi DP angry + SY aNyaPy or just Dy AGPye an There are no other restrictions on the individual ‘w; because borrowing and short-selling are allowed.where m = n-l and the w ,...,¥,”are unconstrained by virtue of the fact that the relation w, = 1 - 51” w, will ensure 1 that the identity constraint in (14) is satisfied, 4. Optimal portfolio and consumption rules: the equations of optimality. The problem of choosing optimal Portfolio and consumption rules for an individual who lives T years is formulated as follows, (5) Max B, [2 u(octy t)ae + a(uen,2)] Subject to: W(o) = Wo the budget con- straint (14), which in the cese of a "risk-free" asset becomes (1'); and where the utility function (during life), U, is assumed to be strictly concave in C and the "bequest" function, B, 4s assumed also to be concave in w. ‘12 To derive the optimal rules, the technique of stochastic dynamic programming ig used. Define (16) 30, P,t) = Max By [2 U(C,s)ds + (ucr),2)] {o,w} where as before, "E," is the conditional expectation operator, conditional on W(t) = W and Py(t) = Py. Define a7) o(w,C5H,P,t) = U(c,t) +£E5] , given wy(t) = wy, C(t) = Cy W(t) = Hy and Py(t) = P tar 1 neve there is no "risk-free" asset, it is assumed no asset can be expressed as a linear combination of the other assets, implying thet the nxn variance covariance matrix of returns, @ = [o,,] where oy 2 po ca ede non-singular. In’the caséJunen therdJig RiN7*°I arisk-free" asset, the same assumption is made about the "reduced" mxm variance-covariance matrix, 43),, pwc 'Z" 4s short for the rigorous £5", , the Dynkin operator over the variables P end W for a given set of controls wand C, f-2 au 3a n a f 22 Wye », Lege +P wey | it Star, b 2yinpin twa? 22 lyanon 4 p a2 +B DIET oyymanyi? Bae + BIE P5955 a? ar Mon po coseFrom the theory of stochastic dynamic programming, the following theorem provides the method for deriving the optimal rules, c* and wS, theoren 1.'1. rp the Py(t) are gen- erated by a strong diffusion process, U is strictly concave in C, and B is con- cave in W, then there exists a set of optimal rules (controls), w* and c*, satisfying })? wy = land J(W,P,T) = B(W,T) and these controls satisfy ° O(C¥ wt su P,t) > O(C,w3W,P,t) for t € [o,T). From theorem I, we have that (18) © = Max {$(C,w3W,P,t)} {C,w) In the usual fashion of maximization under constraint, we define the Lagrangian, L = $ + A[1 Sar wy] where A is the multiplier and find the extreme points from the first- order conditions G3) 0 = Ip(c#,w#) = upcort) - g, 2 = . n eye (20) ° (C# Ww) Ae TyyW + Tu SO oF 5H n 7 FP Tsu eghse KeLse eon (21) oO je where the notation for partial derivatives is Jy au as = 32 Ye = 30 Fa ® apy > Yay F = a ow - For a heuristic proof of this theorem and the derivation of ‘the stochastic Beliman equation, see Dreyfus [li] and Merton [12]. Yor a rigorous proof and discussion of Weaker conditions, sec Kushner [9], chapter IV especially theorem 7.--10- From the theory of stochastic dynamic programming, the following theorem provides the method for deriving the optimal rules, C# and w*. 4) Theorem 1,14 If the Py(t) are gen- erated by a strong diffusion process, U is strictly concave in C, ané B is con- cave in W, then there exists a set of optimal rules (controls), w* and c* satisfying })" wy = 1 and J(W,P,?) = B(W,T) and these controls satisfy © = $(C#,wSW,P,t) > 6(Cyw3¥,P,t) for t € [o,T]. From theorem I, we have that (18) © = Max ($(C,w3W,P,t)} {C,w} In the usual fashion of maximization under constraint, we define the Lagrangian, L = $ + ALl Sys wy] where \ is the multiplier and find the extreme points from the first- order conditions (19) 0 = Lg (c#,w#) = UL(e#,t) - J, Byes " n © ou? (20) ° Tyg OR OM ) At TyoW + Ty, a 5" st n " = FT putes Pgs Kedar on (21) O = L)(c# wt) = 1 ~ Dm" je | where the notation for partial derivatives is J, = 34, J, au arn a For a heuristic proof of this theorem and the derivation of the stochastic Bellman equation, see Dreyfus [4] and Merton [12]. For a rigorous proof and discussion cf Weaker conditions, see Kushner [9], chapter IV especially theorem 7.-te s Because Leg = %¢¢ 7 Upg < % Ug, = , = 05 ke Ke L, ofWFT sb, = 0, k# Jj, a sufficient condition te = OK Tas May for a unique intertor maximum 1s that Jyy <0 (L.e. that J be strictly concave in W), That assumed, as an immediate consequence of differentiating (19) totally with respect to I, we have ace (22) ay? To solve explicitly for C* and w#, we solve the n+2 non-dynamic implicit equations, (19) - (21), for C*, and u¥, and X as functions of Jys Iyys Tyys Ws P and t. Then C# and w* are substituted in (18) which now becomes a second- order partial differential equation for J, subject to the boundary condition J(W,P,T) = B(W,T). Having (in principle at least) solved this equation for J, we then substitute back into (19) - (21) to derive the optimal rules as func- tions of W, P, and t. Define the inverse function @ = (Ug]7"'. Then from (19), (23) ct = a(t) To solve for the wy*, note that (20) is a linear system in w)* and hence can be solved explicitly. Define (24) 9 = Co » the nxn variance-covariance matrix 1) 15) =q7 (vyjlea ngqn PSSO vay. Eliminating 4 from (20), the solution fcr w*, can be written as (25) Lid = hy (Pt) + m(P,W te, (P,t) + f.(PyW,t), K=l,...,n 6) where J" hy = 1, HP a, = 0, and Pi"e, = 0. a6 5)- 15 ot exists by the assumption on Q in footnote 12. 16) ae by. (P, ts Dig? 3 M(PAW,t) 2 ~ T/T yy (continued)meneame aemnamaeeaeemnamriaeeccacmemeaammaaeaammamamana mmm ~12- Substituting for w* and c# in (18), we arrive at the fundamental partial differential equation for J as a function of W, P, and t, ngn MDa Meg! ] (26) O= UlG,t] +I, + Jy [: r -G nm 1 Nn W n +MY FasaPa #2 DEE gagPsPy tr DT ywPy 5 Wn n noon, Figo (Ba Pai ~ BY ywes BP Ba) 2 Tyg a_fan gn n A + oro aig BF sPntmg? Fay?) | a(n Dy Mente? (af WW) Yer%) ‘| subject to the boundary condition J(W,P,T) = B(W,T). If (26) were solved, the solution J could be substituted into (23) and (25) to obtain C# and w* as functions of W, P, and t. For the case where one of the assets is "risk-free", the equations are somewhat simplified because the problem ean be solved directly as an unconstrained maximum by elim- inating w, as was done in (14"), In this case, the optimal Proportions in the risky assets are (27) The partial differential equation for J corr (26) beconies (28) oO " a 6) (continuea) Lyn 8 (Pst) = yy “Sy, pn se oauhs WO vy,subject to the boundary condition J(W,P,?) = B(W,T). Although (28) is a simplified version of (26), neither ; (26) nor (28) lend themselves to easy solution. The com- plexities of (26) and (28) are caused by the basic non- linearity of the equations and the large number of state variables. Although there is little that can be done about the non-linearities, in some cases, it may be possible to reduce the number of state variables. 5. Log-normality of prices and the continuous- time analog to Tobin-Markowitz mean-variance analysis. When, for k=1,...sn, a, ando, are constants, the asset prices have stationary, log-normal distributions. In this case, J will be a function of lf and t only and not P, Then (26) re- duces to qn n > Vig Gy JW 29 O= Ule,t] +o, +9, Wy BY Mest ya]e ane > t Ww T dy? nogn n ~ yy [x Ta Merteeal - (x7 nt va%)* | From (25), the optimal portfolio rule becomes (30) wh = hy + m(W,t)g, where J)? hy From (30), the following "separation" or "mutual fund" 1 and ss &, = 0 and h, and g, are’ constants. theorem can be proved, nn Theorem IT. prices P, whose changes are station- Given n assets with arily and log-normally distributed, then (1) there exist a unique pair (except for scale) of “mutual funds" constructed from linear combinations op 1” see cass and Stiglitz [1] for a general discussion of Separation theorems. The only degenerate case is when | all the assets are identically distributed (i.e. symmetry) in which case, only one mutual fund is needed.-1h- of these assets such that, independent of preferences (i.e. the form.of the utility function) wealth distribution, or time horizon, individuals will be indifferent between choosing from a linear combination of these two funds or a linear combination of the original n assets. (2) If Pp, is the price per share of either fund, then Py is log- normally distributed. Further (3) if §, = péreentage of one mutual fund's value held in the k®® asset end if Ay = Percentage of the other mutual fund's value held in the k*2 asset, then one can find that 8 = Nye t Bey KeLys een and ey =1y.e.an Proof: (1) (30) is @ parametric representation of a line in the hyperplane defined by3} "x My = 1/18 hence, there exist two linearly independent vectors (namely, the vectors of asset proportions held by the two mutuel funds) which form a basis for all optimal portfolios chosen by the individuals. Therefore, each individual would be incifferent between chocsing a linear combination of the mutual fund shares or a linear combination of the original n assets. (2) Let V = NpPp, = the total velue of (either) fund where Np = number of shares of the funé outstanding. Let Ny ge Mky/V = vereentage of total value invested in the KP kB asset. then v =SI" yr, -yn an > aN (31) av = x Nar, + YY Pan, + YH APL aN, number of shares of asset k held ty the fund and and NpaPp + Pydly + aPpdNy fBut 1 (32)? Pang + Tp T4P,aN, = net inflow of funds from non- capital gain. sources = net value of new shares issued = Pde + AN dP From (31) and (32), we have that -y" (33) Np@Pp = SY NAP, By the definition of V and u,, (33) can be re-written as dP. dP, (34) 7 " x u a Mk PR n n = TT bya # TT oy O82, By [t6's Lemma and (34), we have that L . (35) Ppt) = Pplo) onl Hyde — BEET BP wytyeie3) € +O e,] So, Pelt) is log-normally distributed. (3) Let a(W,t;U) = percentage of wealth invested in the first mutual fund by an individual with utility function U and wealth W at time t. Then, (1-a) must equal the per- centage of wealth invested in the second mutual fund. Because the individual is indifferent between these asset holdings or an optimal portfolio chesen from the original n assets, it must be that (36) wh = hy + mW, tle, = ab) + Clea)aAy , kelyes ean Clearly, the only solution to the linear system (36) for all W, t, and U is 7 G37) > ke Lyeeeyn p ke lyeeeynNote that J)" 6 Q.E.D. For the case when one of the assets ig "risk-free", there is a corollary to theorem II. Namely, Corollary. If one ‘of the assets is “pisk-free", then one of the two mutual funds will contain only this asset. If 8, = percentage of the total value of the "risky" mutual fund invested in the k®P asset, then 8 TT Meg (ager TT Diy Mag (agrP)s eLae some Proof. By the assumption of stationary log-normal prices, (27) reduces to (38) Toot Uy Mg (gts KEL and J 2 1- pt = Ww mon . (39) yh dB at phe TY vagy By the same argument as in the proof of theorem II, (38) and (39) define a line in the hyperplane defined by avy =. and -ym nm sm 7 = 2 EDT vg lag-Pd7 ST OM yyy lage) O, k#ly.+.,m ° ken. @.E.D, Thus, if we have an economy where all asset prices are log-normally distributed, the investment decision can be divided into two parts by the establishment of two financial intermediaries (mutual funds) to hold all individual secu- rities and to issue shares of their own for purchase by in~ dividual investor The separation is complete because the "instructions" given the fund managers, namely to hold pro--17- portions 6, and A, of the k&2 security, k=1,...,n, depend only on the price distribution parameters and are independent of individual preferences, wealth distribution, or age distribution. The similarity of this result to that of the classical Tobin-Markowitz analysis is clearest when one examines closely the investment rule given to the "risky" fund's man- ager when there exists a "risk-free" asset (money) with zero return (r=0). It is easy to show that the 6, proportions prescribed in the corollary are derived by finding the locus of points in the (instantaneous) mean-standard deviation space of composite returns which minimize variance for a given mean, and then by finding the point where a line drawn from the origin is tangent to the locus. This point determines the 6, as is illustrated in figure 1. e oe Crocus of ming? ish Fer & given & figure 1. on Given the of, the 6, are determined. So the log-normal assumption in the continuous-time model is sufficient to allow the same analysis as in the static mean-variance model but without the objectionable assumptions of quadratic utility or normality of the distribution of absolute price ob- changes. (Log-normality of price changes is much le: Jectionable, since this does invoke "limited Liability” and by the central limit theorem is the only reguler solution toany continuous-space, infinitely-divisible process in time.) An immediate advantage for the present analysis is that whenever log-normality of prices’ is assumed, we can work, without loss of generality, with Just two assets, one "risk- free" and one risky with its price log-normally distributed. The risky asset can always be thought of as a composite asset with price P(t) defined by the process ap (40) > adt todz where 21h ym mon G2 AEP ys lay-ray/ SH vy s(ay-r) rm Dy 885% 840A /E + Explicit solutions for a particular class of ions. On the assumption of log-normality of prices, some characteristics of the asset demand functions were shown, If a further assumption about the preferences of the individual is made, then equation (28) can be solved in closed form, and the optimal consumption and portfolio rules derived explicitly. Assume that the utility function for the individual, U(C,t), can be written as u(c,t) = eP*y(c) where V is a member of the family of utility functions whose measure of absolute risk aversion is positive and yes in consumption, i.e. A(C) = -W"/V! = 1/ (Gt n/p) >-0, subject to the restrictions (42) y# a> 05 (B24 a) > 05, 022 All members of the HARA (hyperbolic absolute risk-aversion) family can be expressed as (43) v(c) =