A Strategy Which Maximizes the Geometric Mean Return on Portfolio Investments James H. Vander Weide; David W. Peterson; Steven F. Maier Management Science, Vol. 23, No. 10. (Jun., 1977), pp. 1117-1123. Stable URL: http://links.jstor.org/sici?sici=0025-1909%28197706%2923%3A10%3C1117%3AASWMTG%3E2.0.CO%3B2-T Management Science is currently published by INFORMS. Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/informs.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers, and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take advantage of advances in technology. For more information regarding JSTOR, please contact [email protected]. http://www.jstor.org Mon Jun 25 14:39:25 2007 MANAGEMENT SCIENCE Vol. 23, No. 10, June, 1977 Printed in U.S.A. A STRATEGY WHICH MAXIMIZES THE GEOMETRIC MEAN RETURN ON PORTFOLIO INVESTMENTS* JAMES H. VANDER WEIDE?, DAVID W. PETERSON? AND STEVEN F. MAIER? A common formulation of the portfolio selection problem leads to the prescription of a strategy which maximizes the geometric mean return on investments. In this paper we examine conditions under which solutions exist for the case where the returns distribution is discrete. We establish necessary and sufficient conditions for a solution to exist and give a computationally convenient and exact method for finding a solution in circumstances where (i) a solution exists and (ii) the number of securities equals or exceeds the number of values in the returns distribution. 1. Introduction The geometric mean investment strategy, introduced into the finance and economics literature by Henry LatanC [6] in 1959, has recently received some attention in scholarly circles after a decade of near neglect. Most of this work has been devoted to the investigation of various properties of the geometric mean strategy. Among the properties of optimal geometric-mean portfolios recently discovered are (i) they maximize the probability of exceeding a given wealth level in a fixed amount of time, (ii) they minimize the long-run probability of ruin, and (iii) they maximize the expected growth rate of wealth.' Relatively little attention has been given, on the other hand, to either the computational problem of finding the optimal geometricmean portfolio2 or the question of the existence of such a p ~ r t f o l i oIn . ~ this paper, we analyze both of these latter problems under various assumptions about the investor's opportunity set and the form of his subjective probability distribution of holding period returns. We reach two major conclusions. First, optimal geometric-mean portfolios do not exist for some rather obvious formulations of the investor's decision p r ~ b l e mSecond, .~ optimal geometric-mean portfolios are in some circumstances more easily computed than was previously re~ognized.~ * Accepted by Edwin H. Neave; received June 21, 1976. This paper has been with the authors 1 month, for 1 revision. Duke University. For discussion and derivations of these properties see [2], (51, [6], [lo], and [15]. We know of only three or four papers which purport to discuss the computation of optimal geometric mean portfolios. The first, by Elton and Gruber [4], assumes that total portfolio returns are lognormally distributed. But the analytical circumstances under which this would occur seem rather limited, since lognormal security returns do not imply lognormal portfolio returns. The second, by Ziemba [17], discusses the problem and proposes nonlinear programming as a solution means. The third, by Zierdba, Parkan and Brooks-Hill [18], pertains to the problem in which returns are normally distributed. A possible fourth paper is Ziemba's [16], which discusses general approaches to solving optimization problems with stochastic objective functions. Cass and Stiglitz [3] give conditions for a n optimal solution to exist for the case where the returns distribution is finite and the number of securities exactly equals the number of states in the returns distribution. Their result is a special case of our Theorem 1. Leland [7] gves a sufficient condition for a solution to exist, but assumes that there is no direction of recession which offers a nonnegative return for all states of nature. Bertsekas [l] strengthens Leland's result by extending it to a more general class of utility functions and shows that for this class it can be made both necessary and sufficient. Though Bertsekas' approach is applicable to the geometric mean problem, his theorems require some modification to account for the fact that the logarithmic utility function is defined only for positive values of its argument. Richard Roll [13] unwittingly formulates a version of the geometric-mean problem for which a solution is unlikely to exist. Mossin [lo], for instance, says "The main disadvantage-at least at present---of the growth optimal policy as compared e.g., with the selection of an E, S efficient policy, is the formidable computational problem." 1117 ' Copyright 0 1977, The Institute of Management Sciences 1118 J. H. VANDER WEIDE,D. W. PETERSON AND S. F. MAIER 2. The Existence of Optimal Geometric-Mean Portfolios Consider an investor faced with the repetitive problem of investing his wealth in some combination of m alternative investment opportunities. Suppose that the investor's objective is to maximize the geometric mean return on his wealth. If xi denotes the fraction of his wealth the investor invests in opportunity i and r, > 0 denotes the periodic payout, including return of principal, provided by opportunity i, then we may identify the task of finding the optimal allocation of wealth for the coming time period as that of solving the following mathematical programming problem: Maximize E {log x Tr} subject to x Te = 1, (I1 and E is the expectation where x = (x,, . . . , x,)~, r = (r,, . . . , r,JT, e = (1, . . . , operator. As with any mathematical programming problem, there is in connection with (1) not only the question of finding the best value for x, but also the question of whether a best value exists. A solution x* is a finite m-vector which imparts maximal value to the objective function. We discuss the question of whether a solution exists for two particular cases of the geometric mean problem. The General Case The general case of the geometric mean problem is one which imposes no restrictions on the investor's subjective probability distribution of holding period returns. The existence of a solution to a general geometric mean problem can be assured by suitably bounding the set of wealth allocations considered feasible. For example, if to the general problem (1) there are appended the restrictions xi > bi, i = 1, . . . , m, where the bi-are known, finite constants summing to 1 or less, there results the problem we call the Constrained General Problem (CGP). The CGP always has a solution, for it consists of the maximization of a continuous function over a nonempty compact set. We refer to the case where the feasible set is unrestricted as the Unconstrained General Problem (UGP). The UGP has a solution in many circumstances, an important one being that in which the returns ri are jointly distributed in accordance with a probability density function which is positive for all r, combinations near zero; that is P(r) > 0 for all positive r satisfying 0 < rTe < 6, where 6 is a sufficiently small positive n ~ m b e r One . ~ example of such a situation is that in which the logarithm of the return for the ith security is a, PiI E,, where a, and Pi are constants particular to that security, I is a market index, and E, is an independent normal random variable. In such circumstances, if one or more of the x, is negative, there is a positive probability that xTr < 0. Hence any portfolio for which all xi > 0 is at least as desirable, and one may as well append these constraints. But now a CGP results, and as noted previously, such problems have solutions. + + The Discrete Case In the event the r, are discrete random variables which can assume a finite number n of combinations of values, the Unconstrained General Problem may be stated Suppose that for some 6 > 0, the probability measure P governing r assigns positive probability to all open subsets of T = ( r > 0 I r Te < 6 ). Then T is open and nonempty. If some component of x is negative, then the set S = ( r > 0 I rTx > 0) is open and nonempty, and the intersection S n T is open and nonempty. Evidently P ( S ) P(S n T ) > 0, so such an x cannot be strictly preferred to any one with components all of which are nonnegative. STRATEGY TO MAXIMIZE GEOMETRIC MEAN PORTFOLIO RETURN 11 19 equivalently as the following Unconstrained Discrete Problem (UDP): n Maximize 2 [ P, log((x j= 1 T~ ),I subject to x Te = 1. (2) Nature, in this formulation, is capable of assuming any of the states j = 1, . . . , n, and the probability she will assume the jth is denoted 5.The values of the ri, when Nature assumes the jth state, are denoted r,. The m x n matrix of r, values is denoted R and assumed known. One may consider the UDP as an original formulation of the myopic portfolio selection problem, as did Latank [6],or as an approximation to an Unconstrained General Problem such as (1) permitting other than discrete returns r,. This latter view is taken in [8]. If there exists a finite and feasible vector x* which imparts a maximal value to the objective function, then we say x* is a solution to the UDP. The assertion that a solution exists precludes the possibility that the objective function can be made arbitrarily large, but does not preclude its being negatively infinite for all feasible vectors x, as would occur if one or more columns of R were all-zero. To avoid the latter possibility, we now assume that no column of R contains only zeroes. The following theorem gives necessary and sufficient conditions that a solution to the UDP exists. THEOREM 1. The UDP has a solution if and only if there exists an n x 1 vector y, all of whose components are positive, such that Ry = e, where e is the m x 1 vector of ones. PROOF. Suppose there exists y > 0 such that Ry = e, and suppose x satisfies x T~ 1 0, x Te = 1. Let y,, denote the component of y which is minimal, and note that = 1/ymin,hence the feasible region y,, > 0. Then 0 L (x T ~ )5j x T ~ 5e x of ((x T ~ ) ,. ,. . , (XT ~ ) , )values is bounded. Since the region is also closed, it is compact, and the objective function, being continuous on this set, assumes a finite maximum. If there exists no y > 0 such that Ry = e, then there exists by Tucker's Theorem of the Alternative [9, p. 341 an x such that x T~ 2 0, x Te = 0.7 Then ax + e l m is feasible and imparts to the objective function a value which can be made larger than any preassigned number by taking a sufficiently large. Hence, in such a situation, no solution to the UDP exists. The condition Ry = e, y > 0, ensures the objective function assumes a maximal value for some choice of finite xi's, though the set of xi's imparting maximal value to the objective need not be bounded. If the condition fails to hold, then the feasible region is unbounded in such a way that one can make the objective function arbitrarily large by selling short one or more securities and buying others. A test of the condition can be implemented by using a linear programming code to find a feasible solution to Ry = e, y 2 ce, for some small positive c. A geometrical interpretation of the condition Ry = e, y > 0, is that e lies in the interior of the positive cone spanned by the columns of R . The condition of Theorem 1 can be restated in equivalent form: 1'. There exists no solution to the UDP if and only if there exists some x THEOREM such that x T~ 2 0 and x Te = 0. (We use " > " to mean " 2 but not = ".) ' Tucker's Theorem of the Alternative, as given in [9, Table 2.4.11 yields the stated result if we set B' = R, C = 0, D = e T , x = x, and y, = y, where the entities on the left of the equal signs appear in [9], and those on the right are our own symbols. 1120 J. H. VANDER WEIDE, D. W . PETERSON AND S. F. MAIER PROOF.^ Tucker's theorem of the alternative [9]. If x satisfies the condition of Theorem It, it has positive components summing to, say, d > 0, and negative components summing to - d. Denote by x + ( x - ) the vector obtained by setting to zero the negative (positive) components of x . Then x + / d and x - / ( - d ) are each nonnegative vectors whose components sum to 1, and may each be considered a portfolio specification. Furthermore, ( x + x - )=R > 0, SO (X + / d ) = R > - ( x - / d ) = R . Hence Theorem 1' may be restated as, "An infinite solution to the UDP exists if and on& if some convex combination of securities dominates some other (disjoint) convex combination of securities for every state of nature." If an x satisfying the condition of Theorem 1' is available, one can do arbitrarily well by letting the portfolio proportions be a x + e l m , and making a a large number. + 3. The Likelihood of the Existence of a Solution to the Unconstrained Discrete Problem Having established exact circumstances under which the UDP has a solution, we examine next the likelihood with which these circumstances occur. In this regard, it is helpful to consider a special form of the returns distribution in which one security dominates all other securities for each state of nature. In addition, we will assume that for each state of nature all securities besides the dominant one have equal returns. Let a radial vector be defined as any m-component vector r which has nonnegative components, all of which are equal, save one which is greater than the rest. Let the n columns of R be radial vectors selected probabilistically in such a way that the probability the ith component exceeds the other components of the jth column vectors is l / m , 1 2 j 2 n, 1 5 i 5 m. For R to be such that there exists a y > 0 satisfying Ry = e, the columns of R must include at least one radial vector for which the first component is largest, one radial vector for which the second is largest, and so forth. This leads at once to the conclusion that if the n columns of R are radial vectors (no matter how they are chosen) and if n < m , then there is no y > 0 such that Ry = e, and hence no solution to the UDP exists. If n 2 m and the columns of R are chosen as described above, then there is some positive probability that a solution to the UDP exists, and that probability increases with increasing n. For n = m, it happens that the probability of a solution is m ! / m m ,which equals for m = 2 and decreases rapidly with increasing m. For n 1 m , the probability of a solution is + which for m fixed increases toward 1 as n gets larger. Using Monte-Carlo methods, we have shown in an unpublished study that even for more general types of R matrices, the likelihood of the existence of a solution to the UDP increases as the number n of states of nature increases. 4. Solution of the UDP A solution to the Unconstrained Discrete Problem (2) can often be obtained with the aid of an efficient nonlinear programming algorithm. The authors report their experiences in solving geometric-mean portfolio problems with one particularly efficient nonlinear programming code in [8]. As we show below, however, if n does not exceed the number m of securities, it is often possible to obtain optimal alloTheorem 1' may also be obtained as a consequence of Rockafellar's Theorem 27.3 [12, p. 2671 which asserts the nonexistence of a solution to be equivalent to the existence of a common direction of recession for the objective function and the feasible set. STRATEGY TO MAXIMIZE GEOMETRIC MEAN PORTFOLIO RETURN 1121 cations even more efficiently through the solution of a certain set of simultaneous linear equations. Throughout the discussion of these equations, we assume that (i) a solution to the UDP exists and (ii) the payout matrix R has rank n. 2. If H is an m X n matrix such that H ~ R= I,,,then x constitutes a THEOREM solution to the UDP if and only if PROOF. (Follows [14, pp. 77-78], with some modifications and extensions). Necessity. By hypothesis, a solution x* = (x:, . . . , x i ) exists. Since rank(R) = n, no column of R is 0, and so x* imparts a finite (as opposed to negatively. infinite) value to the objective function. Hence X * ~ R > 0. Define the Lagrangian function 2 [ Pj log((x R' I),) n L(x) = - he 'x. j= 1 It is necessary then that x* satisfy VL(x*) = Ry* -Ae = 0, ~ * ~= e 1, , where y * = (y:, . . . , y,*)T, yJ* = P//(x*~R),, j = 1, . . . , n, and A is a scalar constant. Premultiplying (4) by x * and ~ noting (5) and (6), we observe that A = 1. By hypothesis, rank(R) = n so there exists (see Penrose [ll]) an m X n matrix H such ~)~, that H ~ R= I. From (4), y * = HTe. Using this in (6) yields ( x * ~ R=) ~P ~ / ( H ~ and necessity has been established. Sufficiency. Retrace the above steps, bearing in mind that a solution is presumed to exist, that L(x) is concave, and that a saddle point sufficiency condition is applicable. This theorem raises the exciting prospect that the UDP often can be solved through the solution of (3). The latter can be accomplished for fairly large R matrices using a standard linear programming computer code. Even more straightforward is the situation in which R is square and of full rank, for then Theorem 2 becomes: COROLLARY.If rank(R) = n = m, let H T = R unique solution to (1) is given by x = R -IF, where -'. Then the x that constitutes the P/ = P,/(R -le)j, j = 1, . . . , n. The applicability of Theorem 2 is limited, unfortunately, by two elements of its hypothesis. The first of these is that the matrix R have rank n. Since R is of dimension m X n, this restriction requires m > n, or equivalently that the number of securities exceed or equal the number of states of nature. The second limiting element in the hypothesis of Theorem 2 is the assumption that a solution to the UDP exists. As is explained in the previous section, this is unlikely except in cases where n exceeds m or where the xi are further constrained. With the imposition of additional constraints (such as nonnegativity), the development of a theorem analogous to Theorem 2 does not seem possible, and one is apparently forced to retreat to nonlinear programming codes to obtain numerical solutions. EXAMPLE.The following example illustrates a situation where the results of the Corollary to Theorem 2 apply. Suppose that the returns on ten securities in each of ten states of nature are given by the following return matrix: J. H. VANDER WEIDE, D. W . PETERSON AND S. F. MAIER 1 2 3 4 5 States of Nature 6 7 8 9 1 0 Security 1 Security 5 Security 10 In addition, suppose that the probability that each state of nature occurs is given by: Then, since rank(R) = 10 = n = m, the optimal amount to invest in each security is given by the Corollary to Theorem 2. Using this corollary, we have xl=-0.857, x,=14.952, x , = -0.244, x4=0.048, x 5 = -1.284 With this highly unbalanced portfolio we would get an annual return of 66.2%. 5. Conclusion Though much has been written about the desirable properties of growth-optimal portfolios, relatively little attention has been given to the problem of computing the wealth allocation necessar.y to achieve such a portfolio. This latter problem, with which the present paper is concerned, has two major aspects, one dealing with the existence of a solution, and the other with the numerical determination of a solution in those instances where a solution exists. For a wealth allocation model called the Unconstrained General Problem (UGP) we show that there are circumstances under which a solution does not exist, and provide conditions sufficient to ensure that a solution does exist. For this general problem, in cases where a solution exists, it is suggested that recourse to a nonlinear programming code may be necessary to obtain a numerical solution. A special case of the UGP is the Unconstrained Discrete Problem (UDP). For this problem too it is found that a solution need not exist, and that it probably does not exist in the case where the number of securities exceeds the number of possible states of nature. It is, unfortunately, just such a relationship between the numbers of securities and states of nature which permits a solution, if it exists, to be found readily as a solution to a set of simultaneous linear equations. In case the number of states of nature exceeds the number of securities, it is likely that a solution exists. In this case, though, it seems that a nonlinear programming code must be used to compute the solution, a task with which our experience is reported elsewhere [8]., The authors wish to acknowledge Professor Henry Latank's helpful insights, particularly in regard to the dominance characteristic associated with infinite solutions. STRATEGY TO MAXIMIZE GEOMETRIC MEAN PORTFOLIO RETURN 1123 References BERTSEKAS, DIMITRIP., "Necessary and Sufficient Conditions for Existence of an Optimal Portfolio," Journal of Economic Theory, Vol. 8 (1974), pp. 235-247. BREIMAN, LEO, "Investment Policies for Expanding Business Optimal in a Long-Run Sense," Naoal Research Logistics Quarterly, Vol. 7 (December 1960), pp. 647-651. CASS,DAVIDAND STIGLITZ, JOSEPHE. "The Structure of Investor Preferences and Asset Returns, and Separability in Portfolio Allocation: A Contribution to the Pure Theory of Mutual Funds," Journal of Economic Theory, Vol. 2 (1970), pp. 122-160. ELTON,EDWINJ. AND GRUBER,MARTINJ., "On the Maximization of the Geometric Mean with Lognormal Return Distribution," Management Science, Vol. 21 (December 1974), pp. 483488. NILS H., "Capital Growth and the Mean-Variance Approach to Portfolio Selection," HAKANSSON, Journal of Financial and Quantitatioe Analysis, Vol. 6 (January 1971), pp. 517-555. LATANE, HENRYA., "Criteria for Choice Among Risky Ventures," Journal of Political Economy, Vol. 38 (April 1959), pp. 145-155. LELAND,HAYNEE., "On the Existence of Optimal Policies under Uncertainty," Journal of Economic Theory, Vol. 4 (1972), pp. 3 5 4 . F., PETERSON, MAIER,STEVEN DAVIDW. AND VANDER WEIDE,JAMESH., "A Monte Carlo Investigation of Characteristics of Optimal Geometric Mean Portfolios," to appear in the Journal of Financial and Quantitative Analysis (June 1977). MANGASARIAN, OLVIL., Nonlinear Programming, McGraw-Hill Book Company, 1969, Chapter 2. MOSSIN,JAN,Theory of Financial Markets, Prentice-Hall, Inc., Englewood Cliffs, New Jersey, 1973. PENROSE, R., "A Generalized Inverse for Matrices," Proceedings of the Cambridge Philosophical Society, Vol. 51 (July 1955), pp. 406413. R. T., Convex Analysis, Princeton University Press, Princeton, N.J., 1970. ROCKAFELLAR, ROLL, RICHARD,"Evidence on the 'Growth-Optimum' Model," Journal of Finance, Vol. 28 (June 1973), pp. 551-566. W., The Mathematical Theory of Communication, The University of SHANNON, C. E. AND WEAVER, Illinois Press, Urbana, 1964. THORP,EDWARD,"Portfolio Choice and the Kelly Criterion," The L & L Proceedings of the Business and Economic Statistics Section, American Statistical Association, 1971. ZIEMBA, WILLIAM, "Solving Nonlinear Programming Problems with Stochastic Objective Functions," Journal of Financial and Quantitatioe Analysis, Vol. 7 (June 1972), pp. 1809-1828. -, "Note on 'Optimal Growth Portfolios when Yields are Serially Correlated'," Journal of Financial and Quantitatioe Analysis, Vol. 7 (September 1972), pp. 1995-2000. -, PARKAN,C. AND BROOKS-HILL, R., "Calculation of Investment Portfolios with Risk Free Borrowing and Lending," Management Science, Vol. 21 (October 1974), pp. 209-222. http://www.jstor.org LINKED CITATIONS - Page 1 of 3 - You have printed the following article: A Strategy Which Maximizes the Geometric Mean Return on Portfolio Investments James H. Vander Weide; David W. Peterson; Steven F. Maier Management Science, Vol. 23, No. 10. (Jun., 1977), pp. 1117-1123. Stable URL: http://links.jstor.org/sici?sici=0025-1909%28197706%2923%3A10%3C1117%3AASWMTG%3E2.0.CO%3B2-T This article references the following linked citations. If you are trying to access articles from an off-campus location, you may be required to first logon via your library web site to access JSTOR. Please visit your library's website or contact a librarian to learn about options for remote access to JSTOR. [Footnotes] 1 Capital Growth and the Mean-Variance Approach to Portfolio Selection Nils H. Hakansson The Journal of Financial and Quantitative Analysis, Vol. 6, No. 1. (Jan., 1971), pp. 517-557. Stable URL: http://links.jstor.org/sici?sici=0022-1090%28197101%296%3A1%3C517%3ACGATMA%3E2.0.CO%3B2-6 2 On the Maximization of the Geometric Mean with Lognormal Return Distribution Edwin J. Elton; Martin J. Gruber Management Science, Vol. 21, No. 4, Application Series. (Dec., 1974), pp. 483-488. Stable URL: http://links.jstor.org/sici?sici=0025-1909%28197412%2921%3A4%3C483%3AOTMOTG%3E2.0.CO%3B2-K 2 Note on "Optimal Growth Portfolios when Yields are Serially Correlated" William T. Ziemba The Journal of Financial and Quantitative Analysis, Vol. 7, No. 4. (Sep., 1972), pp. 1995-2000. Stable URL: http://links.jstor.org/sici?sici=0022-1090%28197209%297%3A4%3C1995%3ANO%22GPW%3E2.0.CO%3B2-N 2 Calculation of Investment Portfolios with Risk Free Borrowing and Lending W. T. Ziemba; C. Parkan; R. Brooks-Hill Management Science, Vol. 21, No. 2, Application Series. (Oct., 1974), pp. 209-222. Stable URL: http://links.jstor.org/sici?sici=0025-1909%28197410%2921%3A2%3C209%3ACOIPWR%3E2.0.CO%3B2-3 NOTE: The reference numbering from the original has been maintained in this citation list. http://www.jstor.org LINKED CITATIONS - Page 2 of 3 - 2 Solving Nonlinear Programming Problems with Stochastic Objective Functions William T. Ziemba The Journal of Financial and Quantitative Analysis, Vol. 7, No. 3. (Jun., 1972), pp. 1809-1827. Stable URL: http://links.jstor.org/sici?sici=0022-1090%28197206%297%3A3%3C1809%3ASNPPWS%3E2.0.CO%3B2-9 4 Evidence on the "Growth-Optimum" Model Richard Roll The Journal of Finance, Vol. 28, No. 3. (Jun., 1973), pp. 551-566. Stable URL: http://links.jstor.org/sici?sici=0022-1082%28197306%2928%3A3%3C551%3AEOT%22M%3E2.0.CO%3B2-H References 4 On the Maximization of the Geometric Mean with Lognormal Return Distribution Edwin J. Elton; Martin J. Gruber Management Science, Vol. 21, No. 4, Application Series. (Dec., 1974), pp. 483-488. Stable URL: http://links.jstor.org/sici?sici=0025-1909%28197412%2921%3A4%3C483%3AOTMOTG%3E2.0.CO%3B2-K 5 Capital Growth and the Mean-Variance Approach to Portfolio Selection Nils H. Hakansson The Journal of Financial and Quantitative Analysis, Vol. 6, No. 1. (Jan., 1971), pp. 517-557. Stable URL: http://links.jstor.org/sici?sici=0022-1090%28197101%296%3A1%3C517%3ACGATMA%3E2.0.CO%3B2-6 8 A Monte Carlo Investigation of Characteristics of Optimal Geometric Mean Portfolios Steven F. Maier; David W. Peterson; James H. Vander Weide The Journal of Financial and Quantitative Analysis, Vol. 12, No. 2. (Jun., 1977), pp. 215-233. Stable URL: http://links.jstor.org/sici?sici=0022-1090%28197706%2912%3A2%3C215%3AAMCIOC%3E2.0.CO%3B2-V NOTE: The reference numbering from the original has been maintained in this citation list. http://www.jstor.org LINKED CITATIONS - Page 3 of 3 - 13 Evidence on the "Growth-Optimum" Model Richard Roll The Journal of Finance, Vol. 28, No. 3. (Jun., 1973), pp. 551-566. Stable URL: http://links.jstor.org/sici?sici=0022-1082%28197306%2928%3A3%3C551%3AEOT%22M%3E2.0.CO%3B2-H 16 Solving Nonlinear Programming Problems with Stochastic Objective Functions William T. Ziemba The Journal of Financial and Quantitative Analysis, Vol. 7, No. 3. (Jun., 1972), pp. 1809-1827. Stable URL: http://links.jstor.org/sici?sici=0022-1090%28197206%297%3A3%3C1809%3ASNPPWS%3E2.0.CO%3B2-9 17 Note on "Optimal Growth Portfolios when Yields are Serially Correlated" William T. Ziemba The Journal of Financial and Quantitative Analysis, Vol. 7, No. 4. (Sep., 1972), pp. 1995-2000. Stable URL: http://links.jstor.org/sici?sici=0022-1090%28197209%297%3A4%3C1995%3ANO%22GPW%3E2.0.CO%3B2-N 18 Calculation of Investment Portfolios with Risk Free Borrowing and Lending W. T. Ziemba; C. Parkan; R. Brooks-Hill Management Science, Vol. 21, No. 2, Application Series. (Oct., 1974), pp. 209-222. Stable URL: http://links.jstor.org/sici?sici=0025-1909%28197410%2921%3A2%3C209%3ACOIPWR%3E2.0.CO%3B2-3 NOTE: The reference numbering from the original has been maintained in this citation list.
© Copyright 2026 Paperzz