portfolio optimization analyticderivative
-
Upload
endu-wesen -
Category
Documents
-
view
228 -
download
0
Transcript of portfolio optimization analyticderivative
-
8/14/2019 portfolio optimization analyticderivative
1/66
-
8/14/2019 portfolio optimization analyticderivative
2/66
-
8/14/2019 portfolio optimization analyticderivative
3/66
-
8/14/2019 portfolio optimization analyticderivative
4/66
-
8/14/2019 portfolio optimization analyticderivative
5/66
ALFRED P. SLOAN SCHOOL OF MANAGEMEI
AN ANALYTIC DERIVATIONOF THE EFFICIENT PORTFOLIO FRONTIER
493-70Robert C, Merton
October 1970
MASSACHUSETTSINSTITUTE OF TECHNOLOGY
J 50 MEMORIAL DRIVECAMBRIDGE, MASSACHUSETTS 02139
-
8/14/2019 portfolio optimization analyticderivative
6/66
-
8/14/2019 portfolio optimization analyticderivative
7/66
AN ANALYTIC DERIVATIONOF THE EFFICIENT PORTFOLIO FRONTIER
493-70Robert C, Merton
October 1970
-
8/14/2019 portfolio optimization analyticderivative
8/66
-
8/14/2019 portfolio optimization analyticderivative
9/66
AN ANALYTIC DERIVATION OF THE EFFICIEm: PORTFOLIO FRONTIER '
Robert C. MertonMassachusetts Institute of Technology
October 1970
I. Introduction . The characteristics of the efficient (in themean-variance sense) portfolio frontier have been discussed at lengthin the literature. However, for more than three assets, the generalapproach has been to display qualitative results in terms of graphs.In this paper, the efficient portfolio frontiers are derived explicitly,and the characteristics claimed for these frontiers verified. Themost important implication derived from these characteristics, theseparation theorem, is stated andproved in the context of a mutual fundtheorem. It is shown that under certain conditions, the classical
graphical technique for deriving the efficient portfolio frontier isincorrect.
II. The efficient portfolio set when all securities are risky .Suppose there are m risky securities with the expected return on thei^ security denoted by 0< . ; the covariance of returns between the iand j'-^ security denoted by
-
8/14/2019 portfolio optimization analyticderivative
10/66
-
8/14/2019 portfolio optimization analyticderivative
11/66
2risky _, then (T^ > 0, i = 1, . . ., m, and we further assume that nosecurity can be represented as a linear combination of the other se-curitieSj i.e. the variance-covariance matrix of returns, -IL = [ fl - ]is non-singular. The frontier of all feasible portfolios which can beconstructed from these m securities is defined as the locus of feasibleportfolios which have the smallest variance for a prescribed expectedreturn. Let $ . = percentage of the value of a portfolio investedin the i security, i = 1^ . ', ^, and as a definitional result,^-l . = 1. Then, the frontier can be described as the set of port-folios which satisfy the constrained minimization problem,
(1) min |-
-
8/14/2019 portfolio optimization analyticderivative
12/66
O
^
t 5-
-
8/14/2019 portfolio optimization analyticderivative
13/66
where ^ and 2 ^re the multipliers. A critical point occurs wherethe partial derivatives of (2) with respect to S 3 A iiandy^2 ^^^ equal to zero^ i.e.
(3a) = :2'TjO-j- A,0^, -^2, 1 = 1, . . .,
(3b) = c^ - Si ^i
-
8/14/2019 portfolio optimization analyticderivative
14/66
-
8/14/2019 portfolio optimization analyticderivative
15/66
) 2TS. = A^STsTv,jand C > 0.^ Solving (7) for A^ and X 2, ^ fi ^ '^'^^'^
(8) Al = ^^^^-^D
X, = (B - A0 0.^ We can now substitute for A^ and y\ 2 ^(8) into (4) to solve for the proportions of each risky asset held in
the frontier portfolio with expected return Of: namely.
^Because JI-^ is positive definite, 0, hence D > 0.
-
8/14/2019 portfolio optimization analyticderivative
16/66
-
8/14/2019 portfolio optimization analyticderivative
17/66
-
8/14/2019 portfolio optimization analyticderivative
18/66
-
8/14/2019 portfolio optimization analyticderivative
19/66
Figure 1 is a graph of (12) where 5? 2 A/C and ^^ = 1/C are theexpected return and variance of the minimum-variance portfolio . Define
%^ to be the proportion of the minimum-variance portfolio investedin the k*- asset, then from (9),
(14) S^ = i-^ , k = 1, . . ., m.
It is usual to present the frontier in the mean-standard deviationplane instead of the mean-variance plane. From (12) and (13), we havethat
(15) -dO< DOFrom (15), CT is a strictly convex function of
-
8/14/2019 portfolio optimization analyticderivative
20/66
-
8/14/2019 portfolio optimization analyticderivative
21/66
il^iA-re. 1
o< OC
-
8/14/2019 portfolio optimization analyticderivative
22/66
-
8/14/2019 portfolio optimization analyticderivative
23/66
The efficient portfolio frontier (the set of feasible portfolioswhich have the largest expected return for a given standard deviation)is the heavy-lined part of the frontier in Figure 2^ starting withthe minimum-variance portfolio and moving to the North-East. The equa-tion for of as a function of XT along the frontier is
(17) 0(* = Ftr /D(c
-
8/14/2019 portfolio optimization analyticderivative
24/66
-
8/14/2019 portfolio optimization analyticderivative
25/66
r I Q icre ^
-
8/14/2019 portfolio optimization analyticderivative
26/66
-
8/14/2019 portfolio optimization analyticderivative
27/66
To prove theorem I. it is sufficient to show that any portfolio onthe efficient frontier can be attained by a linear combination of twospecific portfolios because an optimal portfolio for any individual(as described in the theorem) will be an efficient portfolio.
Equation (9) describes the proportion of the frontier portfolio^with expected return o^* ^ invested in the k*-^ asset, k = 1, . . ., ra.If we define
,m(19) gk - 2i Vi,j(CO(j - A)/D, k = 1, . . ., m\ = 2i Vj^j(B - AO
-
8/14/2019 portfolio optimization analyticderivative
28/66
-
8/14/2019 portfolio optimization analyticderivative
29/66
-
8/14/2019 portfolio optimization analyticderivative
30/66
-
8/14/2019 portfolio optimization analyticderivative
31/66
10
(24) oe
-
8/14/2019 portfolio optimization analyticderivative
32/66
-
8/14/2019 portfolio optimization analyticderivative
33/66
11
2 = /T n2 _ 9A % L.' J. D v2>(26) (T^, then 0^^ > 0(^ > ky/Q, One could show that the line
Although the paper does not impose general equilibrium market clearingconditions, it is misleading to allow as one of the mutual funds aportfolio which no investor would ever hold long.
-
8/14/2019 portfolio optimization analyticderivative
34/66
-
8/14/2019 portfolio optimization analyticderivative
35/66
' iqure 3.
-
8/14/2019 portfolio optimization analyticderivative
36/66
-
8/14/2019 portfolio optimization analyticderivative
37/66
12
Y^ = A i//C is tangent to the ellipse at the point ('2=0, >* = 0) asdrawn in Figure 3. Therefore^ there do not exist two efficient port-
2folios which are uncorrelated. From (28)^ we have that (TgU ^h ^ '^>which implies that all efficient portfolios are positively correlated.Further, CT^^, = (T^ if and only if Vl = A^Z/C. If >^ = ky/C, then,from (24), 0^5= A/C = t^ which implies that the portfolio held by the fund
2with proportions b is the minimum-variance portfolio with w j^ = 1/Cand bj^ = j^. Because 1/C is the smallest variance of any feasibleportfolio, it must be that, for efficient portfolios, C^jj will besmallest when one of the portfolios is the minimum-variance portfolio.In this case, the portfolio of the other fund will have the character-istics that
(30)
-
8/14/2019 portfolio optimization analyticderivative
38/66
-
8/14/2019 portfolio optimization analyticderivative
39/66
13
a satisfy (30), then
(32a) V= ^^^ ' ^^^
m_ 2?T\i( and the portfolio is efficient. IfR > 0< , then J/ < 0, and the portfolio will be inefficient. If R = 6?y = Qy and equation (31) cannot be satisfied by any frontier portfoliowith finite values of 0( and (T , The implications of these results willbe discussed in the following section.
IV. The efficient portfolio set when one of the assets is risk-les .The previous sections analyzed the case when all the available assetsare risky. In this section, we extend the analysis to include a risk-less
stasset, by keeping the same m risky assets as before and adding a (m+l)asset with a guaranteed return R, In an analogous way to (2) in SectionII, the frontier of all feasible portfolios is determined by solving the
problem:
(33) min[i2'T2T^i5jCr-j+ A [ OT - R - :^1S,^OC^ - K)]] .
Notice that the constraint 2 i ^i = ^ ^^^ *^ appear in (33) becausewe have explicitly substituted for S'j^j^ - ^ ~ 2 1 S ^^s ^^' ^^^$ , . . ., Sm are unconstrained by virtue of the fact that ^^^ canalways be chosen such that S^^^^i = ^ ^^ satisfied. This substitu-tion not only simplifies the analytics of solving (33), but also will
-
8/14/2019 portfolio optimization analyticderivative
40/66
-
8/14/2019 portfolio optimization analyticderivative
41/66
14
provide insight into some results derived later in the paper.The first-order conditions derived from (33) are
(34a) = SiS-jCTTj - ACo^i - R), i = 1, . . ., m
(34b) = 0< - R - Si^iCoTi - R).
Clearly, if Of = R, the frontier portfolio is
-
8/14/2019 portfolio optimization analyticderivative
42/66
-
8/14/2019 portfolio optimization analyticderivative
43/66
15
(38) lo(- r| =0 y Cr2 - 2AR + B ,
is drawn in Figure 4. From (35), (37), and (38), the proportionsof risky assets for the frontier portfolios as a function of 0( are
(39) u = (
-
8/14/2019 portfolio optimization analyticderivative
44/66
-
8/14/2019 portfolio optimization analyticderivative
45/66
F'laure ^
.
-
8/14/2019 portfolio optimization analyticderivative
46/66
-
8/14/2019 portfolio optimization analyticderivative
47/66
16
exists a unique pair of efficiently mutual funds, one con-taining only risky assets and the other only the risk-less
asset, such that all risk-averse individuals, who choosetheir portfolios so as to maximize utility functions dependentonly on the mean and variance of their portfolios, will beindifferent between choosing portfolios from among the ori-ginal m+1 assets or from these two funds, if and only ifR
-
8/14/2019 portfolio optimization analyticderivative
48/66
-
8/14/2019 portfolio optimization analyticderivative
49/66
17
where a^-^ = i - Zj i a^^ and h^-^ = I - Zj i \, Solving (40) forand bj^, we have that
(42) ak = flu^/P
\ = i *l - I) n^/y , k = 1, . . ., mand
2(43) aj^^ = b^^ - (A - RC)/i/(CR^ - 2AR + B)
I'nri-l = 1 - ( 2 - 1)(A - RC)/>/(CR^ - 2AR + B),
Now require that one of the funds (say the one with proportionsb) hold only the risk-less asset (i.e., bj^ = 0, k = 1, . , ,, m andbjjj^j = 1) which is accomplished by choosing 1*2 = lo If it is alsorequired that the other fund hold only risky assets (i.e., a^ri-i = 0),then from (43), i^ = (A - RC)/(Cr2 - 2AR + B). Note that if R = A/C,P = which is not allowed, and as can be seen in (43), in this case,a_^, = b ,, =1. From (42), the two mutual funds are different sincenri-l nri-1 ^ 'b^ = for all k = 1, , . ., m and aj^ jt for some k. HoweverZli a.^ = 0, which means that the risky fund holds a hedged portfolioof long and short positions whose net value is zero. If R > A/C, then// < 0, and the portfolio is inefficient (i.e. 0, and the portfolio is efficient. When R < A/C, the com-position of the efficient risky portfolio is
-
8/14/2019 portfolio optimization analyticderivative
50/66
-
8/14/2019 portfolio optimization analyticderivative
51/66
18
-^^ 'v^.Cof, - R)(44) a^ = ^\y^l , k = 1,K^ (A - RC) ' *Thus, theorem II is proved.
The traditional approach to finding the efficient frontier whenone of the assets is risk-less is to graph the efficient frontier forrisky assets only, and then to draw a line from the intercept tangentto the efficient frontier as illustrated in Figure 5. Suppose thatthe point (ot , 0 ) as drawn in Figure 5 exists. Then one could chooseone mutual fund to be the risk-less asset and the other to be (ty , O^ are displayed inFigures 6 and 7, When R = C , there is no tangency for finite ^ andQ , and the frontier lines (with the risk-less asset included) arethe asyn5>totes of the frontier cuirve for risky assets only. When R > (X ,there is a lower tangency and the efficient frontier lies above theupper asynqjtote. Under no condition can one construct the entirefrontier (with the risk-less security included) by drawing tangentlines to the upper and lower parts of the frontier for risky assets
-
8/14/2019 portfolio optimization analyticderivative
52/66
-
8/14/2019 portfolio optimization analyticderivative
53/66
r q iKre S.
(T
-
8/14/2019 portfolio optimization analyticderivative
54/66
-
8/14/2019 portfolio optimization analyticderivative
55/66
cV
/?=-^
Fi a arc 6.
(T
Fiaw-re I.
cr cr
-
8/14/2019 portfolio optimization analyticderivative
56/66
-
8/14/2019 portfolio optimization analyticderivative
57/66
19
only,-*- The intuitive explanation for this result is that with theintroduction of a risk-less asset, it is possible to select a portfoliowith net non-positive amounts of risky-assets which was not possiblewhen one could only choose among risky assets.
Although for individual portfolio selection, there is no reasonto rule out R > o^ , one could easily show that as a general equili-brium solution with homogeneous expectations. Figure 5 is the onlypossible case with (o< ,0 ), the market portfolio's expected returnand standard deviation. Hence, we have as a necessary condition forequilibrium that R < o< ,
Given that the proportions in the market portfolio must be theMkame as in (44)(i,e, o j^ = aj^, k = 1, . . ,, m where M denotes
for the market portfolio )^ the fundamental result of the capitalasset pricing model, the security market line, can be derived directlyas follows:(45) G-^ = 2 S^ (T^.^ k = 1, . . m
5 21 1 (2T ^ij(< j - '^>Mk / (^ - ^C>* ^'^^^ Q< and an upper tangency (e.g. E. Fama [3], p. 26 and M. Jensen[4], p. 174). In W. Sharpe [8], Chapter 4, the figures appear tohave R = ^ and a double tangency.
-
8/14/2019 portfolio optimization analyticderivative
58/66
-
8/14/2019 portfolio optimization analyticderivative
59/66
20
and(46) crl , ZlU'l^K
^T^i ^ ^i ^>/(^ - ^^>* ^'^' (45)(Ofj^ - R)/(A - RC)
and eliminating (A - RC) by combining (45) and (46), we derive
(TkM(47) ^k ^ = ^ (QfM - R) , k = 1, . . ., m,which is the security market line.
-
8/14/2019 portfolio optimization analyticderivative
60/66
-
8/14/2019 portfolio optimization analyticderivative
61/66
21
References
[1] F, Black, Capital Market Equilibritjin with No Riskless Borrowingor Lending, Financial Note 15A, unpublished, August, 1970,
[2] D, Cass and J, Stiglitz, The Structure of Investor Preferencesand Asset Returns, and Separability in Portfolio Allocation: AContribution to the Pure Theory of Mutual Funds, Cowles Founda-tion Paper, May, 1969,
[3] E. Fama, Risk, Return, and Equilibrium, Report 6831, Universityof Chicago Graduate School of Business, June, 1968,
[4] M, Jensen, Risk, The Pricing of Capital Assets, and the Evalua-tion of Investment Portfolios, Journal of Business. Vol, 42,i^ril, 1969,
[5] J, Lintner, The Valuation of Risk Assets and the Selection ofRisky Investments in Stock Portfolios and Capital Budgets,Review of Economies and Statistics. XLVII, February, 1965,
[6] H, Markowitz, Portfolio Selection; Efficient Diversification ofInvestment. J, Wiley, 1959,
[7] R, Merton, Optimum Consvm5)tion and Portfolio Rules in a Continu-ous-time Wbdel, Working paper #58, Department of Economics,Massachusetts Institute of Technology, August, 1970,
[8] W, Sharpe, Portfolio Theory and Capital Markets. McGraw-Hill,1970
[9 J J. Tobin, The Theory of Portfolio Selection, The Theory ofInterest Rates, eds: F, H, Hahn and F, P, R. Brechling, Macmillan,1965,
-
8/14/2019 portfolio optimization analyticderivative
62/66
-
8/14/2019 portfolio optimization analyticderivative
63/66
-
8/14/2019 portfolio optimization analyticderivative
64/66
-
8/14/2019 portfolio optimization analyticderivative
65/66
-
8/14/2019 portfolio optimization analyticderivative
66/66