Portfolio optimization models and mean-variance spanning

tests
Wei-Peng Chen
*
Department of Finance, Hsih-Shin University, Taiwan
c8!"###$ms8%hinet%net
H&imin Ch&ng
'ra(&ate )nstit&te of Finance, *ationa+ Chiao T&ng University, Taiwan
,eng--& Ho
Department of Finance, *ationa+ Centra+ University, Taiwan
.engy&ho$cc%nc&%e(&%tw
Ts&i-/ing Hs&
'ra(&ate )nstit&te of Finance, *ationa+ Chiao T&ng University, Taiwan
tracy%sh&$msa%hinet%net
Prepared for Handbook of Quantitative Finance and Risk Management
*
Wei-Peng Chen is at the Department of Finance at Shih-Hsin University0 H&imin Ch&ng an( Ts&i-/ing Hs& are at the
'ra(&ate )nstit&te of Finance at the *ationa+ Chiao T&ng University% ,eng--& Ho is at Department of Finance, *ationa+ Centra+
University% 1((ress correspon(ence to H&imin Ch&ng, 'ra(&ate )nstit&te of Finance, *ationa+ Chiao T&ng University, 22
Ta-Hs&eh 3oa(, Hsinch& 422"2, Taiwan0 Te+5 688#-4-"788 e9t%"727"0 Fa95 688#-4-"7448#20% :-mai+5
ch&ngh&i$mai+%nct&%e(&%tw%
)n this chapter we intro(&ce the theory an( the app+ication of comp&ter program of mo(ern
portfo+io theory% The notion of (iversification is age-o+( ;(on<t p&t yo&r eggs in one =as.et>,
o=vio&s+y pre(ates economic theory% However a forma+ mo(e+ showing how to ma.e the most of
the power of (iversification was not (evise( &nti+ ?"8, a feat for which Harry @ar.owitA
event&a++y won *o=e+ PriAe in economics%
@ar.owitA portfo+io shows that as yo& a(( assets to an investment portfo+io the tota+ ris. of that
portfo+io - as meas&re( =y the variance Bor stan(ar( (eviationC of tota+ ret&rn - (ec+ines
contin&o&s+y, =&t the e9pecte( ret&rn of the portfo+io is a weighte( average of the e9pecte( ret&rns
of the in(ivi(&a+ assets% )n other wor(s, =y investing in portfo+ios rather than in in(ivi(&a+ assets,
investors co&+( +ower the tota+ ris. of investing witho&t sacrificing ret&rn%
)n the secon( part we intro(&ce the mean-variance spanning test which fo++ows (irect+y from the
portfo+io optimiAation pro=+em%
INTRODUCTION OF MARKOWITZ PORTFOLIO-SELECTION MODEL
Harry @ar.owitA B?"8, ?"?C (eve+ope( his portfo+io-se+ection techniD&e, which came to =e
ca++e( mo(ern portfo+io theory B@PTC% Prior to @ar.owitA<s wor., sec&rity-se+ection mo(e+s
foc&se( primari+y on the ret&rns generate( =y investment opport&nities% Stan(ar( investment a(vice
was to i(entify those sec&rities that offere( the =est opport&nities for gain with the +east ris. an(
then constr&ct a portfo+io from these% Fo++owing this a(vice, an investor might conc+&(e that
rai+roa( stoc.s a++ offere( goo( ris.-rewar( characteristics an( compi+e a portfo+io entire+y from
these% The @ar.owitA theory retaine( the emphasis on ret&rn0 =&t it e+evate( ris. to a coeD&a+ +eve+
of importance, an( the concept of portfo+io ris. was =orn% Whereas ris. has =een consi(ere( an
important factor an( variance an accepte( way of meas&ring ris., @ar.owitA was the first to c+ear+y
an( rigoro&s+y show how the variance of a portfo+io can =e re(&ce( thro&gh the impact of
(iversification, he propose( that investors foc&s on se+ecting portfo+ios =ase( on their overa++ ris.-
rewar( characteristics instea( of mere+y compi+ing portfo+ios from sec&rities that each in(ivi(&a++y
have attractive ris.-rewar( characteristics%
1 @ar.owitA portfo+io mo(e+ is one where no a((e( (iversification can +ower the portfo+io<s
ris. for a given ret&rn e9pectation Ba+ternate+y, no a((itiona+ e9pecte( ret&rn can =e gaine( witho&t
increasing the ris. of the portfo+ioC% The @ar.owitA :fficient Frontier is the set of a++ portfo+ios of
8
which e9pecte( ret&rns reach the ma9im&m given a certain +eve+ of ris.%
The Markowitz model is =ase( on severa+ ass&mptions concerning the =ehavior of investors
an( financia+ mar.ets5
% 1 pro=a=i+ity (istri=&tion of possi=+e ret&rns over some ho+(ing perio( can =e estimate(
=y investors%
8% )nvestors have sing+e-perio( &ti+ity f&nctions in which they ma9imiAe &ti+ity within the
framewor. of (iminishing margina+ &ti+ity of wea+th%
4% Earia=i+ity a=o&t the possi=+e va+&es of ret&rn is &se( =y investors to meas&re ris.%
!% )nvestors care on+y a=o&t the means an( variance of the ret&rns of their portfo+ios over a
partic&+ar perio(%
"% :9pecte( ret&rn an( ris. as &se( =y investors are meas&re( =y the first two moments of
the pro=a=i+ity (istri=&tion of ret&rns-e9pecte( va+&e an( variance%
#% 3et&rn is (esira=+e0 ris. is to =e avoi(e(

%
7% Financia+ mar.ets are friction+ess%
MEASURMENT OF RETURN AND RISK
Thro&gho&t this chapter, investors are ass&me( to meas&re the +eve+ of ret&rn =y comp&ting the
e9pecte( va+&e of the (istri=&tion, &sing the pro=a=i+ity (istri=&tion of e9pecte( ret&rns for a
portfo+io% 3is. is ass&me( to =e meas&ra=+e =y the varia=i+ity aro&n( the e9pecte( va+&e of the
pro=a=i+ity (istri=&tion of ret&rns% The most accepte( meas&res of this varia=i+ity are the variance
an( stan(ar( (eviation%
Return
'iven any set of ris.y assets an( a set of weights that (escri=e how the portfo+io investment is

@ar.owitA mo(e+ ass&mes that investors are ris. averse% This means that given two assets that offer the same e9pecte(
ret&rn, investors wi++ prefer the +ess ris.y one% Th&s, an investor wi++ ta.e on increase( ris. on+y if compensate( =y
higher e9pecte( ret&rns% Converse+y, an investor who wants higher ret&rns m&st accept more ris.% The e9act tra(e-off
wi++ (iffer =y investor =ase( on in(ivi(&a+ ris. aversion characteristics% The imp+ication is that a rationa+ investor wi++
not invest in a portfo+io if a secon( portfo+io e9ists with a more favora=+e ris.-ret&rn profi+e - i%e%, if for that +eve+ of ris.
an a+ternative portfo+io e9ists which has =etter e9pecte( ret&rns%
Using ris. to+erance, we can simp+e c+assify investors into three types5 ris.-ne&tra+, ris.-averse, an( ris.-+over% 3is.-
ne&tra+ investorFs (o not reD&ire the ris. premi&m for ris. investments0 they G&(ge ris.y prospects so+e+y =y their
e9pecte( rates of ret&rn% 3is.-averse investors are wi++ing to consi(er on+y ris.-free or spec&+ative prospects with
positive premi&m0 they ma.e investment accor(ing the ris.-ret&rn tra(e-off% 1 ris.-+over is wi++ing to engage in fair
games an( gam=+es0 this investor a(G&sts the e9pecte( ret&rn &pwar( to ta.e into acco&nt the Ff&nF of confronting the
prospectFs ris.%
4
sp+it, the genera+ form&+as of e9pecte( ret&rn for n assets is5
( )

B C
n
P i i
i
E r wE r

(X.1)
where5

n
i
i
w

H %20
n H the n&m=er of sec&rities0
i
w
H the proportion of the f&n(s investe( in sec&rity i0
,
i P
r r
H the ret&rn on ith sec&rity an( portfo+io p0 an(
( ) E
H the e9pectation of the varia=+e in the parentheses%
The ret&rn comp&tation is nothing more than fin(ing the weighte( average ret&rn of the
sec&rities inc+&(e( in the portfo+io%
Risk
The variance of a sing+e sec&rity is the e9pecte( va+&e of the s&m of the sD&are( (eviations
from the mean, an( the stan(ar( (eviation is the sD&are root of the variance% The variance of a
portfo+io com=ination of sec&rities is eD&a+ to the weighte( average covariance
2
of the ret&rns on
its in(ivi(&a+ sec&rities5
( ) ( )
8

Ear Cov ,
n n
p p i j i j
i j
r ww r r


BI%8C
Covariance can a+so =e e9presse( in terms of the corre+ation coefficient as fo++ows5
( )
Cov ,
i j ij i j ij
r r
BI%4C
where
ij

H corre+ation coefficient =etween the rates of ret&rn on sec&rity i,

i
r
, an( the rates of ret&rn
on sec&rity j,
j
r
, an(
i

, an(
j

represent stan(ar( (eviations of

i
r
an(
j
r
respective+y% Therefore5
( )

Ear
n n
p i j ij i j
i j
r ww

BI%!C
Jvera++, the estimate of the mean ret&rn for each sec&rity is its average va+&e in the samp+e
perio(0 the estimate of variance is the average va+&e of the sD&are( (eviations aro&n( the samp+e
8
High covariance in(icates that an increase in one stoc.<s ret&rn is +i.e+y to correspon( to an increase in the other% 1
+ow covariance means the ret&rn rates are re+ative+y in(epen(ent an( a negative covariance means that an increase in
one stoc.<s ret&rn is +i.e+y to correspon( to a (ecrease in the other%
!
average0 the estimate of the covariance is the average va+&e of the cross-pro(&ct of (eviations%
EFFICIENT PORTFOLIO
Efficient portfolios may contain any n&m=er of asset com=inations% We e9amine efficient
asset a++ocation =y &sing two ris.y assets for e9amp+e% 1fter we &n(erstan( the properties of
portfo+ios forme( =y mi9ing two ris.y assets, it wi++ =e easy to see how portfo+io of many ris.y
assets might =est =e constr&cte(%
Two-risky-ssets !ort"o#io
Keca&se we now envision forming a -portfo+io from two ris.y assets, we nee( to &n(erstan( how
the &ncertainties of asset ret&rns interact% )t t&rns o&t that the .ey (eterminant of portfo+io ris. i( the
e9tent to which the ret&rns on the two assets ten( to vary rather in tan(em or in opposition% The
(egree to which a two-ris.y-assets portfo+io re(&ces variance of ret&rns (epen(s on the (egree of
corre+ation =etween the ret&rns of the sec&rities%
S&ppose a proportion (enote( =y
A
w
is investe( in asset 1, an( the remain(er

A
w
, (enote( =y
B
w
, is investe( in asset K% The e9pecte( rate of ret&rn on the portfo+io is a weighte( average of the
e9pecte( ret&rns on the component assets, with the same portfo+io proportions as weights%

B C B C B C
P A A B B
E r w E r w E r +
BI%"C
The variance of the rate of ret&rn on the two-asset portfo+io is
8 8 8 8 8 8
B C 8
P A A B B A A B B A B AB A B
w w w w w w + + + BI%#C
where
AB

is the corre+ation coefficient =etween the ret&rns on asset 1 an( asset K% )f the
corre+ation =etween the component assets is sma++ or negative, this wi++ re(&ce portfo+io ris.%
First, ass&me that
%2
AB

, which wo&+( mean that 1sset 1 an( K are perfect+y positive+y
corre+ate(, the right-han( si(e of eqation X.! is a perfect sD&are an( simp+ifies to
8 8 8 8 8
8
8
B C
p A A B B A B A B
A A B B
w w w w
w w


+ +
+
or
"
p A A B B
w w +
Therefore, the portfo+io stan(ar( (eviation is a weighte( average of the component sec&rity
stan(ar( (eviations on+y in the specia+ case of perfect positive corre+ation% )n this circ&mstance,
there are no gains to =e ha( form (iversification% Whatever the proportions of asset 1 an( asset K,
=oth the portfo+io mean an( the stan(ar( (eviation are simp+e weighte( averages% Fig&re I% shows
the opport&nity set with perfect positive corre+ation - a straight +ine thro&gh the component assets%
*o portfo+io can =e (iscar(e( as inefficient in this case, an( the choice among portfo+ios (epen(s
on+y on ris. preference% Diversification in the case of perfect positive corre+ation is not effective%
"igre X.1 #nvestment opportnit$ sets for asset % and asset & wit' varios correlation coefficients
(
Perfect positive corre+ation is the on+y case in which there is no =enefit from (iversification%
With any corre+ation coefficient +ess than %2B
<
C, there wi++ =e a (iversification effect, the
portfo+io stan(ar( (eviation is +ess than the weighte( average of the stan(ar( (eviations of the
component sec&rities% Therefore, there are =enefits to (iversification whenever asset ret&rns are +ess
than perfect+y corre+ate(%
J&r ana+ysis has range( from very attractive (iversification =enefits B
2
AB
<
C to no =enefits at
a++
%2
AB

% For
AB

within this range, the =enefits wi++ =e somewhere in =etween%

*egative corre+ation =etween a pair of assets is a+so possi=+e% Where negative corre+ation is
present, there wi++ =e even greater (iversification =enefits% 1gain, +et &s start with an e9treme% With
4
The proofs of the s+ope an( the shape of e9treme corre+ation =etween asset 1 an( asset K are in 1ppen(i9 1%
2
)tandard *eviation
E
+
p
e
c
t
e
d

,
e
t

r
n
asset K
asset 1

2

2 < <

#
perfect negative corre+ation, we s&=stit&te
%2
AB

in eD&ation I%# an( simp+ify it in the same
way as with positive perfect corre+ation% Here, too, we can comp+ete the sD&are, this time, however,
with (ifferent res&+ts%
8 8
B C
P A A B B
w w
1n(, therefore,
L M
P A A B B
ABS w w BI%7C
With perfect negative corre+ation, the =enefits from (iversification stretch to the +imit%
:D&ation I%7 points to the proportions that wi++ re(&ce the portfo+io stan(ar( (eviation a++ the way
to Aero%
1n investor can re(&ce portfo+io ris. simp+y =y ho+(ing instr&ments which are not perfect+y
corre+ate(% )n other wor(s, investors can re(&ce their e9pos&re to in(ivi(&a+ asset ris. =y ho+(ing a
(iversifie( portfo+io of assets% Diversification wi++ a++ow for the same portfo+io ret&rn with re(&ce(
ris.%
T$e %on%e!t o" Mrkowit& e""i%ient "rontier
:very possi=+e asset com=ination can =e p+otte( in ris.-ret&rn space, an( the co++ection of a++
s&ch possi=+e portfo+ios (efines a region in this space% The +ine a+ong the &pper e(ge of this region
is .nown as the efficient frontier% Com=inations a+ong this +ine represent portfo+ios Be9p+icit+y
e9c+&(ing the ris.-free a+ternativeC for which there is +owest ris. for a given +eve+ of ret&rn%
Converse+y, for a given amo&nt of ris., the portfo+io +ying on the efficient frontier represents the
com=ination offering the =est possi=+e ret&rn% @athematica++y the efficient frontier is the
intersection of the set of portfo+ios with minim&m variance an( the set of portfo+ios with ma9im&m
ret&rn%
Fig&re I%8 shows investors the entire investment opport&nity set, which is the set of a++
attaina=+e com=inations of ris. an( ret&rn offere( =y portfo+ios forme( =y asset 1 an( asset K in
(iffering proportions% The c&rve passing thro&gh 1 an( K shows the ris.-ret&rn com=inations of a++
the portfo+ios that can =e forme( =y com=ining those two assets% )nvestors (esire portfo+ios that +ie
to the northwest in Fig&re I%8% These are portfo+ios with high e9pecte( ret&rns Btowar( the north of
the fig&reC an( +ow vo+ati+ity Bto the westC%
7
"igre X.2 #nvestment opportnit$ set for asset % and asset &
The area within c&rve '(AZ is the feasi=+e opport&nity set representing a++ possi=+e portfo+io
com=inations% Portfo+ios that +ie =e+ow the minim&m-variance portfo+io Bpoint EC on the fig&re can
therefore =e reGecte( o&t of han( as inefficient% The portfo+ios that +ie on the frontier (A in Fig&re
I%8wo&+( not =e +i.e+y can(i(ates for investors to ho+(% Keca&se they (o not meet the criteria of
ma9imiAing e9pecte( ret&rn for a given +eve+ of ris. or minimiAing ris. for a given +eve+ of ret&rn%
This is easi+y seen =y comparing the portfo+io represente( =y points K an( KF% Since investors
a+ways prefer more e9pecte( ret&rn than +ess for a given +eve+ of ris., KF is a+ways =etter than K%
Using simi+ar reasoning, investors wo&+( a+ways prefer K to E =eca&se it has =oth a higher ret&rn
an( a +ower +eve+ of ris.% )n fact, the portfo+io at point E is i(entifie( as the minimm-variance
portfolio0 since no other portfo+io e9ists that has a +ower stan(ar( (eviation% The c&rve (A
represents a++ possi=+e efficient portfo+ios an( is the efficient frontier
!
, which represents the set of
portfo+ios that offers the highest possi=+e e9pecte( rate of ret&rn for each +eve+ of portfo+io stan(ar(
(eviation%
!
The efficient frontier wi++ =e conve9 N this is =eca&se the ris.-ret&rn characteristics of a portfo+io change in a non-
+inear fashion as its component weightings are change(% B1s (escri=e( a=ove, portfo+io ris. is a f&nction of the
corre+ation of the component assets, an( th&s changes in a non-+inear fashion as the weighting of component assets
changes%C The efficient frontier is a para=o+a Bhyper=o+aC when e9pecte( ret&rn is p+otte( against variance Bstan(ar(
(eviationC%
1
@inim&m
Eariance
Portfo+io
B@EPC
2
K
KF
)nvestment
Jpport&nity
Set
E
)tandard *eviation
E
+
p
e
c
t
e
d

,
e
t

r
n

O
8
"igre X.( -'e efficient frontier of risk$ assets and individal assets
1ny portfo+io on the (own war( s+oping potion of the frontier c&rve is (ominate( =y the
portfo+io that +ies (irect+y a=ove it on the &pwar( s+oping portion of the frontier c&rve since that
portfo+io has higher e9pecte( ret&rn an( eD&a+ stan(ar( (eviation% The =est choice among the
portfo+ios on the &pwar( s+oping portion of the frontier c&rve is not as o=vio&s, =eca&se in this
region higher e9pecte( ret&rn is accompanie( =y higher ris.% The =est choice wi++ (epen( on the
investorFs wi++ingness to tra(e off ris. against e9pecte( ret&rn%
S$ort se##in)
Eario&s constraints may prec+&(e a partic&+ar investor from choosing portfo+ios on the efficient
frontier, however% Short sa+e restrictions are on+y one possi=+e constraint% Short sa+e is a &s&a+
reg&+ate( type of mar.et transaction% )t invo+ves se++ing assets that are =orrowe( in e9pectation of a
fa++ in the assetsF price% When an( if the price (ec+ines, the investor =&ys an eD&iva+ent n&m=er of
assets at the new +ower price an( ret&rns to the +en(er the assets that was =orrowe(%
*ow, re+a9ing the ass&mption of no short se++ing, investors co&+( se++ the +owest-ret&rn asset K
Bhere, we ass&me that
B C B C an(
A B A B
E r E r
C% )f the n&m=er of short sa+es is &nrestricte(,
then =y a contin&o&s short se++ing of K an( reinvesting in 1 the investor co&+( generate an infinite
e9pecte( ret&rn% The efficient frontier of &nconstraint portfo+io is shown in Fig&re I%!% The &pper
=o&n( of the highest-ret&rn portfo+io wo&+( no +onger =e 1 =&t infinity Bshown =y the arrow on the
top of the efficient frontierC% /i.ewise the investor co&+( short se++ the highest-ret&rn sec&rity 1 an(
2
:fficient frontier
of ris.y assets
)tandard *eviation
E
+
p
e
c
t
e
d

,
e
t

r
n
)n(ivi(&a+ assets
@inim&m
variance
portfo+io
?
reinvest the procee(s into the +owest-yie+( sec&rity K
"
, there=y generating a ret&rn +ess than the
ret&rn on the +owest-ret&rn assets% 'iven no restriction on the amo&nt of short se++ing, an infinite+y
negative ret&rn can =e achieve(, there=y removing the +ower =o&n( of K on the efficient frontier%
Hence, short se++ing genera++y wi++ increase the range of a+ternative investments from the minim&m-
variance portfo+io to p+&s or min&s infinity
#
%
"igre X.. -'e efficient frontier of nrestricted/restricted portfolio
3e+a9ing the ass&mption of no short se++ing in this (eve+opment of the efficient frontier
invo+ves a mo(ification of the ana+ysis of the efficient frontier of constraint Bnot a++owe( short
sa+esC% *e9t section, we intro(&ce the mathematica+ ana+ysis of the efficient frontier withPwitho&t
short se++ing constraints%
C#%u#tin) t$e Mini*u* +rin%e !ort"o#io
)n @ar.owitA portfo+io mo(e+, we ass&me investors choose portfo+ios =ase( on =oth e9pecte(
ret&rn,
B C
p
E r
, an( the stan(ar( (eviation of ret&rn as a meas&re of its ris.,
p

% So, the portfo+io

se+ection pro=+em can =e e9presse( as ma9imiAing the ret&rn with respect to the ris. of the
investment Bor, a+ternative+y, minimiAing the ris. with respect to a given ret&rn, ho+( the ret&rn
constant an( so+ve for the weighting factors that minimiAe the varianceC%
"
3ationa+ investor wi++ not short se++ a high-ret&rn asset an( =&y a +ow-ret&rn asset% This case is G&st for e9treme
ass&mption%
#
Whether an investor engages in any of this short-se++ing activity (epen(s on the investorFs own &niD&e set of
in(ifference c&rves%
2
OF
O
1
P
E
E
+
p
e
c
t
e
d

,
e
t

r
n
)tandard *eviation
K
witho&t short sa+es
with short sa+es
2
@athematica++y, the portfo+io se+ection pro=+em can =e form&+ate( as D&a(ratic program% For
two ris.y assets 1 an( K, the portfo+io consists of
,
A B
w w
, the ret&rn of the portfo+io is then, The
weights sho&+( =e chosen so that Bfor e9amp+eC the ris. is minimiAe(, that is
8 8 8 8 8
@in 8
A
P A A B B A B AB A B
w
w w w w + +
for each chosen ret&rn an( s&=Gect to
, 2, 2
A B A B
w w w w +
% The +ast two constraints
simp+y imp+y that the assets cannot =e in short positions%
The minim&m variance portfo+io weights are shown in Ta=+e I%, the (etai+ proofs are in
1ppen(i9 K%
-a0le X.1 -'e mimimm variance portfolio weig't of two-assets portfolio wit'ot s'ort selling
The corre+ation of two assets Weight of 1sset 1 Weight of 1sset K
QH
B
A
A B
w

8
A B
B
A B
w

QH -
B
A
A B
w

+
A
B
A B
w

+
QH 2
8
8 8
B
A B
A
w

+
8 8
8 8
8
A B
A B
B
w

+
1=ove, we simp+y &se two-ris.y-assets portfo+io to ca+c&+ate the minim&m variance portfo+io
weights% )f we genera+iAation to portfo+ios containing assets, the minim&m portfo+io weights can
then =e o=taine( =y minimiAing the /agrange f&nction ! for portfo+io variance%
8

@in
n n
p i j ij i j
i j
ww

S&=Gect to
8
%%%

w w w + + +
( )

Cov
n n n
i j i j i
i j i
! ww rr "

_
+

,

BI%8C
in which

are the /agrange m&+tip+iers, respective+y,

ij

is the corre+ation coefficient =etween

i
r

an(
j
r
, an( other varia=+es are as previo&s+y (efine(%
Ky &sing this approach the minim&m variance can =e comp&te( for any given +eve+ of e9pecte(
portfo+io ret&rn Bs&=Gect to the other constraint that the weights s&m to oneC% )n practice it is =est to
&se a comp&ter =eca&se of the e9p+osive increase in the n&m=er of ca+c&+ations as the n&m=er of

sec&rities consi(ere( grows% The efficient set that is generate( =y the aforementione( approach
BeD&ation I%8C is sometimes ca++e( the minimm-variance set =eca&se of the minimiAing nat&re of
the /agrangian so+&tion%
C#%u#tin) t$e wei)$ts o" o!ti*# risky !ort"o#io
Jne of the goa+s of portfo+io ana+ysis is minimiAing the ris. or variance of the portfo+io%
Previo&s section intro(&ce the ca+c&+ation of minim&m variance portfo+io, we minim&m the
variance of portfo+io s&=Gect to the portfo+io weightsF s&mming to one% )f we a(( a con(ition into
the eD&ation I%8# whish is =e s&=Gect to the portfo+ioFs attaining some target e9pecte( rate of
ret&rn, we can get the optima+ ris.y portfo+io%
8

@in
n n
p i j ij i j
i j
""

S&=Gect to
( )
*

n
i i
i
" E R E

, where
*
E
is the target e9pecte( ret&rn an(

%2
n
i
i
"

The first constraint simp+y says that the e9pecte( ret&rn on the portfo+io sho&+( eD&a+ the target
ret&rn (etermine( =y the portfo+io manager% The secon( constraint says that the weights of the
sec&rities investe( in the portfo+io m&st s&m to one%
The /agrangian o=Gective f&nction can =e written5
( ) ( )
*
8

Cov
n n n n
i j i j i i i
i j i i
! ww rr E wE r w

1 _
+ +
1
] ,

BI%?C
Ta.ing the partia+ (erivatives of this eD&ation with respect to each of the varia=+es,
8
, ,%%%%,

w w w
8
, , an( setting the res&+ting eD&ations eD&a+ to Aero yie+(s the minimiAation of ris.
s&=Gect to the /agrangian constraints% Then, we can so+ve the weights an( these weights are
represente( optima+ ris.y portfo+io =y &sing of matri9 a+ge=ra%
)f there no short se++ing constraint in the portfo+io ana+ysis, secon( constraint,

%2
n
i
i
w

, sho&+(
8
s&=stit&te to

%2
n
i
i
w

, where the a=so+&te va+&e of the weights

i
w
a++ows for a given
i
w
to =e
negative Bso+( shortC =&t maintains the reD&irement that a++ f&n(s are investe( or their s&m eD&a+s
one%
The /agrangian f&nction is
( ) ( )
*
8

Cov
n n n n
i j i j i i i
i j i i
! ww rr E wE r w

1 _
+ +
1
] ,

BI%2C
)f the restriction of no short se++ing is in minimiAation variance pro=+em, it nee(s to a(( a thir(
constraint5
2, , ,
i
w i K
The a((ition of this non-negativity constraint prec+&(es negative va+&es for the weights Bthat
is, no short se++ingC% The pro=+em now is a D&a(ratic programming pro=+em simi+ar to the ones
so+ve( so far, e9cept that the optima+ portfo+io may fa++ in an &nfeasi=+e region% )n this circ&mstance
the ne9t =est optima+ portfo+io is e+ecte( that meets a++ of the constraints%
Fin,in) t$e e""i%ient "rontier o" risky ssets
1ccor(ing to two-f&n( separation, the efficient frontier of ris.y assets can =e forme( =y any
two ris.y portfo+ios one the frontier% 1++ portfo+ios on the mean-variance efficient frontier can =e
forme( as a weighte( average of any two portfo+ios or f&n(s on the efficient frontier1 is ca++e( two-
f&n( separation% So if we have any two points of the portfo+io com=inations, we can (raw an entire
efficient frontier of the ris.y assets% Previo&s sections we have intro(&ce( the minim&m variance
portfo+io an( optima+ ris.y portfo+io given the e9pecte( ret&rn, then, we can generate the entire
efficient frontier =y the separation property%
Deriving the efficient frontier may =e D&ite (iffic&+t concept&a++y, =&t comp&ting an( graphing it
with any n&m=er of assets an( any set of constraints is D&ite straightforwar(% /ater, we wi++ &se
:IC:/ an( @1T/1K to generate the efficient frontier%
Fin,in) t$e o!ti*# risky !ort"o#io
We a+rea(y have an efficient frontier, however, how we (eici(e the =est a++ocation of portfo+ioR
Jne of the factors to consi(er when se+ecting the optima+ portfo+io for a partic&+ar investor is (egree
4
of ris. aversion, investorFs wi++ingness to tra(e off ris. against e9pecte( ret&rn% This +eve+ of
aversion to ris. can =e characteriAe( =y (efining the investorFs in(ifference c&rve, consisting of the
fami+y of ris.Pret&rn pairs (efining the tra(e-off =etween the e9pecte( ret&rn an( the ris.% )t
esta=+ishes the increment in ret&rn that a partic&+ar investor wi++ reD&ire in or(er to ma.e an
increment in ris. worthwhi+e% The optima+ portfo+io a+ong the efficient frontier is not &niD&e with
this mo(e+ an( (epen(s &pon the ris.Pret&rn tra(eoff &ti+ity f&nction of each investor% We &se the
&ti+ity f&nction that is common+y emp+oye( =y financia+ theorists an( the 1)@3 B1ssociation of
)nvestment @anagement an( 3esearchC assigns a portfo+io with a given e9pecte( ret&rn
B C
p
E r
an(
stan(ar( (eviation
p

the fo++owing &ti+ity f&nction5

8
B C 2%22"
p p
$ E r A
BI%C
where $ is the &ti+ity va+&e an( 1 is an in(e9 of the investorFs ris. aversion% The factor of 2%22" is
a sca+ing convention that a++ows &s to e9press the e9pecte( ret&rn an( stan(ar( (eviation in
eD&ation I% as percentages rather than (ecima+s% We interpret this e9pression to say that the
&ti+ity from a portfo+io increases as the e9pecte( rate of ret&rn increases, an( it (ecreases when the
variance increases% The re+ative magnit&(e of these changes is governe( =y the coefficient of ris.
aversion, 1 % For ris.-ne&tra+ investors, 1H2% Higher +eve+s of ris. aversion are ref+ecte( in +arger
va+&es for 1%
Portfo+io se+ection, then, is (etermine( =y p+otting investorsF &ti+ity f&nctions together with the
efficient-frontier set of avai+a=+e investment opport&nities% )n Fig&re I%#, two sets of in(ifference
c&rves +a=e+e(

$
an(
8
$
are shown together with the efficient frontier% The

$
c&rve has a higher
s+ope, in(icating a greater +eve+ of ris. aversion% The investor is in(ifferent to any com=ination of
p
r
an(
p

a+ong a given c&rve% The

8
$
c&rve wo&+( =e appropriate for a +ess ris.-averse investorSthat
is, one who wo&+( =e wi++ing to accept re+ative+y higher ris. to o=tain higher +eve+s of ret&rn% The
optima+ portfo+io wo&+( =e the one that provi(es the highest &ti+itySa point in the northwest
(irection Bhigher ret&rn an( +ower ris.C% This point wi++ =e at the tangent of a &ti+ity c&rve an( the
efficient frontier% :ach investor is +ogica++y se+ecting the optima+ portfo+io given his or her ris.-
ret&rn preference, an( neither is more correct than the other%
!
"igre X.! #ndifference 2rves and Efficient frontier
)n or(er to simp+ify the (etermination of optima+ ris.y portfo+io, we &se the capita+ a++ocation
+ine BC1/C, which (epicts a++ feasi=+e ris.-ret&rn com=inations avai+a=+e from (ifferent asset
a++ocation choices, to (etermine the optima+ ris.y portfo+io% To start, however, we wi++ (emonstrate
the so+&tion of the portfo+io constr&ction pro=+em with on+y two ris.y assets Bin o&r e9amp+e, asset
1 an( asset KC an( a ris.-free asset% )n this case, we can (erive an e9p+icit form&+a for the weights
of each asset in the optima+ portfo+io% This wi++ ma.e it easy to i++&strate some of the genera+ iss&es
pertaining to portfo+io optimiAation%
The o=Gective is to fin( the weights
A
w
an(
B
w
that res&+t in the highest s+ope of the C1/
B i%e%, the weights that res&+t in the ris.y portfo+io with the highest rewar(-to-varia=i+ity ratioC%
Therefore, the o=Gective is to ma9imiAe the s+ope of the C1/ for any possi=+e portfo+io, P% Th&s o&t
o=Gective f&nction is the s+ope that we have ca++e(
S
!A%
5 The entire portfo+io inc+&(ing ris.y an(
ris.-free assets%
B C
p f
S
p
E r r
!A%

BI%8C
For the portfo+io with two ris.y assets, the e9pecte( ret&rn an( stan(ar( (eviation of portfo+io
S are
B C B C B C
P A A B B
E r w E r w E r +
BI%"C
2
K
1
)tandard *eviation
E
+
p
e
c
t
e
d

,
e
t

r
n
@inim&m
variance
portfo+io :fficient
frontier of
ris.y assets

$
8
$
"
8 8 8 8 P 8
B 8 C
P A A B B A B AB A B
w w w w + + BI%4C
When we ma9imiAe the o=Gection f&nction,
S
!A%
, we have to satisfy the constraint that the
portfo+io weights s&m to % Therefore, we so+ve a mathematica+ pro=+em forma++y written as
B C

i
p f
S
w
p
E r r
!A%
Ma&

S&=Gect to

i
w

%)n this case of two ris.y assets, the so+&tion for the weights of the o!ti*#
risky !ort"o#io S, can =e shown to =e as fo++ows
7
5
8
8 8
L B C M L B C M
L B C M L B C M L B C B C M

A f B B f AB A B
A
A f B B f A A f B f AB A B
B A
E r r E r r
w
E r r E r r E r r E r r
w w

+ +

Then, we form an optima+ comp+ete portfo+io
8
given an optima+ ris.y portfo+io an( the C1/
generate( =y a com=ination of portfo+io S an( ris.-free asset% We have constr&cte( the optima+
portfo+io S, we can &se the in(ivi(&a+ investorFs (egree of ris. aversion, 1, to ca+c&+ate the optima+
proportion of comp+ete portfo+io to invest in the ris.y component%
1ss&ming that a ris.-free rate is
f
r
, an( a ris.y portfo+io with e9pecte( ret&rn
B C
p
E r
an(
stan(ar( (eviation
p

wi++ fin( that, for any choice of

'
, the e9pecte( ret&rn of the comp+ete
portfo+io is
B C L B C M
! f p f
E r r ' E r r +
The variance of the overa++ portfo+io is
8 8 8
! p
'
The investor attempts to ma9im&m &ti+ity, U, =y choosing the =est a++ocation to the ris.y asset,
'
% To so+ve the &ti+ity ma9imiAation pro=+em more genera++y, we write the pro=+em as fo++ows5
8
y
8 8
UH B C 2%22"
L B C M 2%22"
@a9 ! !
f p f p
E r A
r ' E r r A'

+
7
The so+&tion proce(&re for two ris.y assets is as fo++ows% S&=stit&te for e9pecte( ret&rn from eD&ationI%" an( for
stan(ar( (eviation from eD&ation I%4% S&=stit&te
A
w for
B
w % Differentiate the res&+ting e9pression for Sp with
respect to
A
w , set the (erivative eD&a+ to Aero, an( so+ve for
B
w %
8
The comp+ete portfo+io means that the entire portfo+io inc+&(ing ris.y an( ris.-free assets%
#
Setting the (erivative of this e9pression to Aero, we can so+ve for
'
yie+( the optima+ position
for ris.-averse investors in the ris.y asset,
*
' , as fo++ows5
*
8
B C
2%2
p f
p
E r r
'
A

BI%!C
The so+&tion shows that the optima+ position in the ris.y asset is, as one wo&+( e9pect,
inverse+y proportiona+ to the +eve+ of ris. aversion an( the +eve+ of ris. Bmeas&re( =y the varianceC
an( (irect+y proportiona+ to the ris. premi&m offere( =y the ris.y asset%
Jnce we have reache( this point, genera+iAing to the case of many ris.y assets is
straightforwar(% Kefore we move on, +et &s =rief+y s&mmariAe the steps we fo++owe( to arrive at the
comp+ete portfo+io%
% Specify the ret&rn characteristics of a++ sec&rities Be9pecte( ret&rns, variances,
covariancesC%
8% :sta=+ish the ris.y portfo+io5
a% Ca+c&+ate the optima+ ris.y portfo+io S%
=% Ca+c&+ate the properties of portfo+io S &sing the weights (etermine( in step an(
eD&ations I%" an( I%4%
4% 1++ocation f&n(s =etween the ris.y portfo+io an( the ris.-free asset5
a% Ca+c&+ate the fraction of the comp+ete portfo+io a++ocate( to Portfo+io S Bthe ris.y
portfo+ioC an( to ris.-free asset BeD&ation I%!C%
=% Ca+c&+ate the share of the comp+ete portfo+io investe( in each asset an( in ris.-free
asset%
7
"igre X.3 *etermination of t'e optimal portfolio
)n practice, when we try to constr&ct optima+ ris.y portfo+ios from more than two ris.y assets
we nee( to re+y on @icrosoft :IC:/ or another comp&ter program% We present can =e &se( to
constr&ct efficient portfo+ios of many assets in the ne9t section%
ALTERNATI(E COMPUTER PRO-AME TO CALCULATE EFFICIENT FRONTIER
Severa+ software pac.ages can =e &se( to generate the efficient frontier% )n this section, we wi++
(emonstrate the metho( &sing @icrosoft :9ce+ an( @1T/1K%
A!!#i%tion. Mi%roso"t E/%e#
:9ce+ is far from the =est program for generating the efficient frontier an( is +imite( in the
n&m=er of assets it can han(+e, =&t wor.ing thro&gh a simp+e portfo+io optimiAer in :9ce+ can
i++&strate concrete+y the nat&re of the ca+c&+ations &se( in more sophisticate( T=+ac.-=o9F programs%
-o& wi++ fin( that :9ce+, the comp&tation o the efficient frontier is fair+y easy%
1ss&me an 1merican investor who forms a si9-stoc.-in(e9 portfo+io% The portfo+io consists of
si9 stoc. in(e9es5 Unite( State BSUP"22C, Unite( ,ing(om BFTS:22C, SwitAer+an BSwiss @ar.et
)n(e9, S@)C ,Singapore BStraits Times )n(e9, ST)C, Hong,ong BHang Seng )n(e9, HS)C , an( ,orea
B,orea Composite Stoc. Price )n(e9, ,JSP)C, with month+y price (ata from Van% ??2 to Dec%
822#% HePshe wants to .now hisPher optima+ portfo+io a++ocation%
)n(ifference C&rve
Jpport&nity Set of 3is.y
1ssets
Jptima+
Comp+ete
Portfo+io
C1/
2
Jptima+ 3is.y Portfo+io
)tandard *eviation
E
+
p
e
c
t
e
d

,
e
t

r
n

f
r
8
The @ar.owitA portfo+io se+ection pro=+em can =e (ivi(e( into three parts% First, we nee( to
ca+c&+ate the efficient frontier% Secon(+y, we nee( to choose the optima+ ris.y portfo+io given oneFs
capita+ a++ocation +ine Bfin( the point at the tangent of a C1/ an( the efficient frontierC% Fina++y,
&sing the optima+ comp+ete portfo+io a++ocate f&n(s =etween the ris.y portfo+io an( the ris.-free
asset%
)tep one4 "inding efficient frontier
First, we nee( to ca+c&+ate e9pecte( ret&rn, stan(ar( (eviation, an( covariance matri9% The
e9pecte( ret&rn an( stan(ar( (eviation can =een ca+c&+ate( =y app+ying the :9ce+ STD:E an(
1E:31': f&nctions to the historic month+y percentage ret&rns (ata
?
% Ta=+e I%81 an( K shows
average ret&rns, stan(ar( (eviations, an( the corre+ation matri9
2
for the rates of ret&rn on the stoc.
in(e9% 1fter we inp&t Ta=+e I%8A into o&r sprea(sheet as shown, we create the covariance matri9 in
Ta=+e I%8B &sing the re+ationship
B , C
i j ij i j
!ov r r
%
-a0le X.2 Performance of si+ stock inde+es
?
The e9pecte( ret&rn an( stan(ar( (eviation of each in(e9 nee( to =e ann&a+iAe(%
2
The corre+ation matri9 is ca+c&+ate( in :9ce+ &sing the (ata ana+ysis f&nction that is fo&n( &n(er the Too+ @en&% *ote
if Data 1na+ysis (oes not appear on the Too+ @en& yo& wi++ nee( to se+ect 1((-in an( a(( to the @en&%
?
-a0le X.2 (2onclded)
82
-a0le X.2 (2onclded)
8
Kefore comp&ting of the efficient frontier, we nee( to prepare the (ata to esta=+ish a
=enchmar. against which to eva+&ate o&r efficient portfo+ios, we can form a =or(er-m&+tip+ie(
covariance matri9% We &se the target mean of "W for e9amp+e% To comp&te the target portfo+ioFs
mean an( variance, these weights are entere( in the =or(er co(umn B)*+B,) an( =or(er row C)-+
H)-% We ca+c&+ate the variance of this portfo+io in ce++ C"# in Ta=+e I%8./ The entry in C"# eD&a+s
the s&m of a++ e+ements in the =or(er-m&+tip+ie( covariance matri9 where each e+ement is first
m&+tip+ie( =y the portfo+io weights given in =oth the row an( co+&mn =or(ers% We a+so inc+&(e two
ce++s to comp&te the stan(ar( (eviation an( e9pecte( ret&rn of the target portfo+io Bform&+as in ce++s
C"7, C"8C

%
To comp&te points a+ong the efficient frontier we &se the :9ce+ So+ver in Ta=+e I%8D Bwhich yo&
can fin( in the Too+s men&C
8
% Jnce yo& =ring &p So+ver, yo& are as.e( to enter the ce++ of the target
Bo=GectiveC f&nction% )n o&r app+ication, the target is the variance of the portfo+io, given in ce++ C"#%
So+ver wi++ minimiAe this target% -o& ne9t m&st inp&t the ce++ range if the (ecision varia=+es B in this
case, the portfo+io weights, containe( in ce++s K!?-K"!C% Fina++y, yo& enter a++ necessary constraints
into the So+ver% For an &nrestricte( efficient frontier that a++ows short sa+es, there are two
constraints5 first, that the s&m of the weights%2 Bce++ K""HC, an( secon(, that the portfo+io
e9pecte( ret&rn eD&a+s target ret&rn "W Bce++ K"8H"C
4
% Jnce yo& have entere( the two
constraints yo& as. the So+ver to fin( the optima+ portfo+io weights%
The So+ver =eeps when it has fo&n( a so+&tion an( a&tomatica++y a+ters the portfo+io weight
ce++s in row !8 an( co+&mn C to show the ma.e&p of the efficient portfo+io% )t a(G&sts the entries in
the =or(er-m&+tip+ie( covariance matri9 to ref+ect the m&+tip+ication =y these new weights, an( it
shows the mean an( variance of this optima+ portfo+io-the minim&m variance portfo+io with mean
ret&rn of "W% These res&+ts are shown in Ta=+e I%8., ce++s C"#-C"8% -o& can fin( that they yie+(
an e9pecte( ret&rn of "W with a stan(ar( (eviation of 7%W Bres&+ts in ce++s C"8 an( C"7C% To
generate the entire efficient frontier, .eep changing the reD&ire( mean in the constraint Bce++ C"8C,

Keca&se the e9pecte( ret&rns an( the portfo+io weights are represente( =y co+&mn vectors B(enote(
e
an(
w
respective+y, with row vector transposes
0
e
an(
0
w
C, an( the variance-covariance terms =y matri9
1
, then the
e9pressions can =e written as simp+e matri9 form&+as% So, the ca+c&+ation of e9pecte( ret&rn an( variance of portfo+io
can &se the :9ce+ array f&nctions%
@atri9 notation :9ce+ form&+a5
Portfo+io ret&rn5
0
w e
HSU@P3JDUCTBw,eC
Portfo+io variance5
0
w 1w
H@@U/TBT31*SPJS:B C,@@U/TB , CC w 1 w
8
)f So+ver (oes not show &p &n(er the Too+s men&, yo& sho&+( se+ect 1((-)ns an( then se+ect 1na+ysis% This sho&+( ass
So+ver to the +ist of options in the Too+s men&%
4
)f yo& (o not set the secon( constraint5 target mean eD&a+ to 82, then yo& can the minim&m variance portfo+io% The
minim&m variance portfo+io has 8%#47W e9pecte( ret&rn an( a stan(ar( (eviation of "%#!4W%
88
+etting the So+ver wor. for yo&% )f yo& recor( a s&fficient n&m=er of points, yo& wi++ =e a=+e to
generate a graph of the D&a+ity of Fig&re I%8%
)f short se++ing is not a++owe(, the So+ver a+so a++ows yo& to a++ ;no short sa+es> an( other
constrains easi+y% We nee( to impose the a((itiona+ constraints that each weight Bthe e+ements in
co+&mn K an( row !?C m&st =e nonnegative% Jnce they are entere(, yo& repeat the variance-
minimiAation e9ercise &nti+ yo& generate the entire restricte( frontier% The o&ter frontier in Figure
2/- is (rawn ass&ming that the investor may maintain negative portfo+io weights, the insi(e frontier
o=taine( a++owing short sa+es% Ta=+e I%8 : an( F present a n&m=er of points on the two frontiers
with an( witho&t short sa+es% -o& can see that the weights in restricte( portfo+ios are never negative%
The minim&m variance portfo+ios in two frontiers are not the same%
Kefore we move on, +et &s s&mmariAe the steps of &sing So+ver to ca+c&+ate the variance-
minimiAation portfo+io%
The steps with So+ver are5
% )nvo.e So+ver =y choosing Too+s then Jptions then So+ver%
8% Specify in the So+ver parameter Dia+og Ko95 the Target ce++ to =e optimiAe( specify ma9
or min
4% Choose 1(( to specify the constrains then J,
!% So+ve an( get the res&+ts in the sprea(sheet%
"igre X.5 Efficient frontier of nrestricted and restricted portfolio
84
Portfolio Efficient "rontier
2
"
2
"
82
8"
42
2 " 2 " 82 8" 42 4"
Portfo+io 3is. BWC
P
o
r
t
f
o
+
i
o

3
e
t
&
r
n

B
W
C
3estricte( Unrestricte( SUP "22 FTS:22 S@) ST) HS) ,JSP)
)tep two4 "inding optimal risk$ portfolio
*ow that we have the efficient frontier, we procee( to step two% )n or(er to get the optima+ ris.y
portfo+io, we sho&+( fin( the portfo+io on the tangency point of capita+ a++ocation an( efficient
frontier% To (o so, we can &se the So+ver to he+p &s% First, yo& enter the of the target f&nction,
;ma9im&m> the rewar(-to-varia=i+ity ratio B
B C
p f
p
E r r

,we ass&me ris.-free rate is !%2W

!
C the
s+ope of the C1/, inp&t the ce++ range Bthe portfo+io weights, containe( in ce++s K!?-K"!C, an( other
necessary constraintsB s&ch +i.e the s&m of the weights eD&a+ to one an( othersC% Then as. the
So+ver to fin( the optima+ portfo+io weights% The res&+ts are shown in Ta=+e I%8 : an( F% The
optima+ ris.y portfo+io with short se++ing a++owance has e9pecte( ret&rn of ?%78!W with a stan(ar(
(eviation of 88%?7#W Bce++ K#?, C#?C% The e9pecte( ret&rn an( stan(ar( (eviation of the restricte(
optima+ ris.y portfo+io are 8%8?7Wan( #%??8W Bce++ K82, C82C%
)tep t'ree4 2apital allocation decision
JneFs a++ocation (ecision wi++ inf+&ence =y his (egree of ris. aversion% *ow we have optima+
ris.y portfo+io, we can &se the concept of comp+ete portfo+io a++ocation f&n(s =etween ris.y
portfo+io an( ris.-free asset% We &se eD&ation I% as o&r &ti+ity f&nction an( set the ris. aversion
eD&a+ to " an( ris.-free rate is !%W% First we constr&ct a comp+ete portfo+io with ris.-free asset an(
!
We &se three month Treas&ry Ki++ interest rate as the ris.-free rate, the average interest rate of 4month T-Ki++ is !%2?W
form ??2P2 to 822#P8%
8!
optima+ ris.y portfo+io%
1ccor(ing to eD&ation I%!, the optima+ weight in ris.y portfo+io is
8
B C
2%2
p f
p
E r r
A

an( the optima+

position of ris.-free asset is -
8
B C
2%2
p f
p
E r r
A

% Then we can &se eD&ation " an( # ca+c&+ate the e9pecte(

ret&rn an( stan(ar( (eviation of the overa++ optima+ portfo+io% The res&+ts are shown in Ta=+e I%8 '
an( H/ The optima+ &nrestricte( Brestricte(C portfo+io has 4%!!W B?%!8WC e9pecte( ret&rn with
4%74W B2%!#WC stan(ar( (eviation% 1n( the investor wi++ invest #2W B#8WC of portfo+io va+&e in
ris.y portfo+io an( !2W B48WC in ris.-free asset%
To s&m &p the three steps, the a++ res&+ts are shown in Ta=+e I%4%
-a0le X.( -'e reslts of optimization pro0lem
portfo+io Unrestricte( portfo+io 3estricte( portfo+io
@inim&m
variance
portfo+io
Jptima+
ris.y
portfo+io
Jptima+
Jvera++
portfo+io
@inim&m
variance
portfo+io
Jptima+
ris.y
portfo+io
Jptima+
Jvera++
portfo+io
Portfo+io
3et&rn
7%74W ?%78W 4%!!W 8%88W 8%8?W ?%!8W
Portfo+io
3is.
8%7W 88%?8W 4%74W 8%84W #%??W 2%!#W
A!!#i%tion. MATLA'
Ky &sing the So+ve, yo& can repeat the variance-minimiAation e9ercise &nti+ yo& generate the
entire efficient frontier, G&st with on p&sh of a =&tton% The comp&tation of the efficient frontier is
fair+y easy in @icrosoft :9ce+% However, :9ce+ is +imite( in the n&m=er of assets it can han(+e% )f
o&r portfo+io consists of h&n(re(s of asset, &sing :9ce+ to (ea+ with the @ar.owitA optima+ pro=+em
wi++ =ecome more comp+icate(% @1T/1K can han(+e this pro=+em% The Financia+ Too+=o9 in
@1T/1K provi(es a comp+ete integrate( comp&ting environment for financia+ ana+ysis an(
engineering% The too+=o9 has everything yo& nee( to perform mathematica+ an( statistica+ ana+ysis
of financia+ (ata an( (isp+ay the res&+ts with presentation-D&a+ity graphics% -o& (o not nee( to
switch too+s, convert fi+es, or rewrite app+ications% V&st write e9pressions the way yo& thin. of
pro=+ems, @1T/1K wi++ (o a++ that for yo&% -o& can D&ic.+y as., vis&a+iAe, an( answer
comp+icate( D&estions%
Financia+ Too+=o9 inc+&(es a set of portfo+io optimiAation f&nctions (esigne( to fin( the portfo+io
8"
that =est meets investor reD&irements% 3eca++ the steps when &sing :9ce+ in previo&s section% )n the
portfo+io optimiAation pro=+em, we nee( to prepare severa+ the (ata for the comp&tation preparation%
First, we nee( the e9pecte( ret&rn, stan(ar( (eviation, covariance matri9% Then, we can ca+c&+ate
the optima+ ris.y portfo+io weights an( .now how a++ocate wi++ =e more efficient% Kefore ta.ing rea+
portfo+io se+ection e9amp+e =y &sing @1T/1K, we intro(&ce the severa+ portfo+io optimiAation
f&nctions that wi++ =e &se in o&r portfo+io se+ection ana+ysis%
Men-+rin%e e""i%ient "rontier
Ca++ing f&nction frontcon can ret&rns the mean-variance efficient frontier for a given gro&p
of assets with &ser-specifie( asset constraints, covariance, an( ret&rns% The comp&tation is =ase( on
sets of constraints representing the ma9im&m an( minim&m weights for each asset, an( the
ma9im&m an( minim&m tota+ weight for specifie( gro&ps of assets% The efficient frontier
comp&tation f&nctions reD&ire information a=o&t each asset in the portfo+io% This (ata is entere( into
the f&nction via two matrices5 an e9pecte( ret&rn vector an( a covariance matri9% Ca++ing
frontcon whi+e specifying the o&tp&t arg&ments ret&rns the correspon(ing vectors an( arrays
representing the ris., ret&rn, an( weights for each of the 2 points comp&te( a+ong the efficient
frontier% Since there are no constraints, yo& can ca++ frontcon (irect+y with the (ata yo& a+rea(y
have% )f yo& ca++ frontcon witho&t specifying any o&tp&t arg&ments, yo& get a graph representing
the efficient frontier c&rve%
"nction4 frontcon
Synta9
[PortRisk, PortReturn, PortWts] = frontcon(ExpReturn, ExpCovariance, NumPorts,
PortReturn, AssetBounds, roups,
roupBounds!
PortRisk, PortReturn an( PortWts are the ret&rns of frontcon, where PortRisk is a
vector of the stan(ar( (eviation of each portfo+io, PortReturn is a vector of the e9pecte( ret&rn
of each portfo+io, an( PortWts is an matri9 of weights a++ocate( to each asset
"
% The o&tp&t (ata is
represente( row-wise, where each portfo+ioFs ris., rate of ret&rn, an( associate( weight is i(entifie(
as correspon(ing rows in the vectors an( matri9%
ExpReturn, ExpCovariance, NumPorts are frontcon reD&ire( information, NumPorts,
PortReturn, AssetBounds , an( roupBounds a++ are optiona+ information% ExpReturn
"
The tota+ of a++ weights in a portfo+io is %
8#
specifies the e9pecte( ret&rn of each asset0 ExpCovariance specifies the covariance of the asset
ret&rns% PortWts is a he matri9 of weights a++ocate( to each asset, an optiona+ item% )n
NumPorts, yo& can (efine the n&m=er of portfo+ios yo& want to =e generate( a+ong the efficient
frontier% )f NumPorts is empty Bentere( as LMC, frontcon comp&tes 2 eD&a++y space( points
#
%
PortReturn is vector of +ength eD&a+ to the n&m=er of portfo+ios containing the target ret&rn
va+&es on the frontier% )f PortReturn is not entere( or LM, NumPorts eD&a++y space( ret&rns
=etween the minim&m an( ma9im&m possi=+e va+&es are &se(% AssetBounds is a matri9
containing the +ower an( &pper =o&n(s on the weight a++ocate( to each asset in the portfo+io
7
%
'ro&ps is n&m=er of gro&ps matri9 specifying asset gro&ps or c+asses
8
% roupBounds matri9
a++ows yo& to specify an &pper an( +ower =o&n( for each gro&p% :ach row in this matri9 represents
a gro&p% The first co+&mn represents the minim&m a++ocation, an( the secon( co+&mn represents the
ma9im&m a++ocation to each gro&p%
*otice, the arg&ments in the parentheses are the f&nction reD&ire( (ata0 the items in the sD&are
=rac.et are the ret&rns of f&nction, the ret&rn wi++ =e show in the o&tp&t% )f yo& (o not want to
res&+ts show &p in o&tp&t win(ow, yo& can c+ear the sD&are =rac.et G&st .eep the item% Jn the
contrary, if yo& want show some res&+ts in the o&tp&t win(ow, G&st a(( sD&are =rac.et when
e9pression fee(=ac. of f&nctions in writing synta9% @oreover, the names of inp&t arg&ments can =e
&ser-specifie(, it is not necessary to =e name( as o&r e9amp+e shown% -o& can name what yo& want,
as +ong as the concept of (ata matches the f&nction arg&ments reD&ire(%
Port"o#ios on %onstrine, e""i%ient "rontier
)f yo& want set some +inear constraints when comp&ting efficient frontier, yo& can ca++
portopt f&nction% The portopt comp&tes portfo+ios a+ong the efficient frontier for a given
gro&p of assets, =ase( on a set of &ser-specifie( +inear constraints% Typica++y, these constraints are
generate( &sing the constraint specification f&nctions which we wi++ (escri=e in ne9t section%
"nction4 portopt
)$nta+
[PortRisk, PortReturn, PortWts] = portopt(ExpReturn, ExpCovariance, NumPorts,
PortReturn, Con"et!
#
When entering a target rate of ret&rn BPortReturnC, enter NumPorts as an empty matri9 LM%
7
The Defa&+t of +ower asset =o&n( is a++ 2s Bno short-se++ingC0 (efa&+t &pper asset =o&n( is a++ s Bany asset may
constit&te the entire portfo+ioC%
8
roupsBi,GC H BGth asset =e+ongs in the ith gro&pC% roupsBi,GC H 2 BGth asset not a mem=er of the ith gro&pC%
87
The portopt is an a(vance( version of frontcon f&nction% The most (ifference =etween
this two is Con"et, others are the same as o&r (escri=e( in frontcon f&nction% Con"et, optiona+
information, is a constraint matri9 for a portfo+io of asset investments, create( &sing portcons% )f
not specifie(, a (efa&+t is create(% The synta9 e9presse( a=ove wi++ ret&rn the mean-variance
efficient frontier with &ser-specifie( covariance, ret&rns, an( asset constraints BCon"etC% )f
portopt is invo.e( witho&t o&tp&t arg&ments, it ret&rns a p+ot of the efficient frontier%
Port"o#io %onstrints
Whi+e frontcon a++ows yo& to enter a fi9e( set of constraints re+ate( to minim&m an(
ma9im&m va+&es for gro&ps an( in(ivi(&a+ assets, yo& often nee( to specify a +arger an( more
genera+ set of constraints when fin(ing the optima+ ris.y portfo+io% The f&nction portopt a((resses
this nee(, =y maccepting an ar=itrary set of constraints as an inp&t matri9% The a&9i+iary f&nction
portcons can =e &se( to create the matri9 of constraints, with each row representing an ineD&a+ity%
These ineD&a+ities are of the type 1*Wts< XH =, where 1 is a matri9, = is a vector, an( Wts is a row
vector of asset a++ocations% The n&m=er of co+&mns of the matri9 1, an( the +ength of the vector Wts
correspon( to the n&m=er of assets% The n&m=er of rows of the matri9 1, an( the +ength of vector =
correspon( to the n&m=er of constraints% This metho( a++ows yo& to specify any n&m=er of +inear
ineD&a+ities to the f&nction portopt
?
%
"nction4 portcons
)$nta+
Con"et = portcons(varar#in
82
!
Con"et H portconsB<Const$%pe<, &ata', %%%, &ataNC creates a matri9 Con"et, =ase( on
the constraint type Const$%pe, an( the constraint parameters &ata', %%%, &ataN(
Con"et H portconsB<Const$%pe'<, &ata'', %%%, &ata'N,<Const$%pe)<, &ata)', %%%,
&ata)N, %%%C creates a matri9 Con"et, =ase( on the constraint types Const$%peN, an( the
correspon(ing constraint parameters &ataN', %%%, &ataNN%
?
)n rea+ity, portcons is an entry point to a set of f&nctions that generate matrices for specific types of constraints%
portcons a++ows yo& to specify a++ the constraints (ata at once, whi+e the specific portfo+io constraint f&nctions a++ow
yo& to =&i+( the constraints incrementa++y% These constraint f&nctions are pcpva*, pca*ims, pc#*ims, an( pc#comp%
82
varar#in is varia=+e +ength inp&t arg&ment +ist% The varar#in statement is &se( on+y insi(e a f&nction @-fi+e to
contain optiona+ inp&t arg&ments passe( to the f&nction% The varar#in arg&ment m&st =e (ec+are( as the +ast inp&t
arg&ment to a f&nction, co++ecting a++ the inp&ts from that point onwar(s% )n the (ec+aration, varar#in m&st =e
+owercase%
88
-a0le X.. -'e detail information of portcons fnction
Constraint type Description Ea+&e
&efau*t
1++ a++ocations are YH 20 no
short se++ing a++owe(%
Com=ine( va+&e of
portfo+io a++ocations
norma+iAe( to
NumAssets BreD&ire(C% Sca+ar representing
n&m=er of assets in portfo+io%
Port+a*ue
Fi9 tota+ va+&e of portfo+io
to PEa+%
P+a* BreD&ire(C% Sca+ar representing tota+ va+&e
of portfo+io%
*&m1ssets BreD&ire(C% Sca+ar representing
n&m=er of assets in portfo+io%
Asset,ims
@inim&m an( ma9im&m
a++ocation per asset%
Asset-in BreD&ire(C% Sca+ar or vector of +ength
*1SS:TS, specifying minim&m a++ocation per
asset%
Asset-ax BreD&ire(C% Sca+ar or vector of +ength
*1SS:TS, specifying ma9im&m a++ocation per
asset%
NumAssets Boptiona+C%
roup,ims
@inim&m an( ma9im&m
a++ocations to asset gro&p%
roups BreD&ire(C% *'3JUPS-=y-*1SS:TS
matri9 specifying which assets =e+ong to each
gro&p%
roup-in BreD&ire(C% Sca+ar or a vector of
+ength *'3JUPS, specifying minim&m
com=ine( a++ocations in each gro&p%
roup-ax BreD&ire(C% Sca+ar or a vector of
+ength *'3JUPS, specifying ma9im&m
com=ine( a++ocations in each gro&p%
roupComparison
'ro&p-to-gro&p
comparison constraints%
roupA BreD&ire(C% *'3JUPS-=y-*1SS:TS
matri9 specifying first gro&p in the comparison%
AtoBmin BreD&ire(C% Sca+ar or vector of +ength
*'3JUPS specifying minim&m ratios of
a++ocations in 'ro&p1 to a++ocations in 'ro&pK%
AtoBmax BreD&ire(C% Sca+ar or vector of +ength
*'3JUPS specifying ma9im&m ratios of
a++ocations in 'ro&p1 to a++ocations in 'ro&pK%
roupB BreD&ire(C% *'3JUPS-=y-*1SS:TS
matri9 specifying secon( gro&p in the
comparison%
Custom
C&stom +inear ineD&a+ity
constraints 1*PortWts< XH
=%
1 BreD&ire(C% *CJ*ST31)*TS-=y-*1SS:TS
matri9, specifying weights for each asset in each
ineD&a+ity eD&ation%
= BreD&ire(C% Eector of +ength *CJ*ST31)*TS
specifying the right han( si(es of the
ineD&a+ities%
So&rce5 @1T/1K Financia+ too+=o9
O!ti*# %!it# ##o%tion to e""i%ient "rontier !ort"o#ios
8?
1 +ast f&nction we (escri=e here may =e &se( to fin( an optima+ portfo+io% So far, we have (ea+
with efficient portfo+ios, +eaving the ris.Pret&rn tra(e-off &nreso+ve(% We may reso+ve this tra(e-off
=y +in.ing mean-variance portfo+io theory to the more genera+ &ti+ity theory% F&nction porta**oc
comp&tes the optima+ ris.y portfo+io on the efficient frontier, =ase( on the ris.-free rate, the
=orrowing rate, an( the investorFs (egree of ris. aversion% 1+so generates the capita+ a++ocation +ine,
which provi(es the optima+ a++ocation of f&n(s =etween the ris.y portfo+io an( the ris.-free asset% )n
the Financia+ too+=o9 the f&nction porta**oc is provi(e(, which yie+(s the optima+ portfo+io
ass&ming the D&a(ratic &ti+ity f&nction5
8
B C 2%22"
p p
$ E r A
, where the parameter 1 is +in.e( to
ris. aversion0 its (efa&+t va+&e is 4%
"nction4 porta**oc
)$nta+
[Risk%Risk, Risk%Return, Risk%Wts, Risk%.raction, /vera**Risk, /vera**Return] =
porta**oc(PortRisk, PortReturn, PortWts, Risk*essRate, Borro0Rate,
RiskAversion!
Risk%Risk is the stan(ar( (eviation of the optima+ ris.y portfo+io0 Risk%Return is the
e9pecte( ret&rn of the optima+ ris.y portfo+io0 Risk%Wts is a vector of weights a++ocate( to the
optima+ ris.y portfo+io0 Risk%.raction is the fraction of the comp+ete portfo+io a++ocate( to the
ris.y portfo+io0 /vera**Risk is the stan(ar( (eviation of the optima+ overa++ portfo+io0
/vera**Return is the e9pecte( rate of ret&rn of the optima+ overa++ portfo+io%
PortRisk is stan(ar( (eviation of each ris.y asset efficient frontier portfo+io0 PortReturn
is e9pecte( ret&rn of each ris.y asset efficient frontier portfo+io0 PortWts is weights a++ocate( to
each asset% Risk*essRate is ris.-free +en(ing rate Bfor investingC, a (ecima+ n&m=er0
Borro0Rate is =orrowing rate, which is +arger than the ris.+ess rate, a (ecima+ n&m=er
8
%
RiskAversion is coefficient of investor<s (egree of ris. aversion BDefa&+t H 4C% Borro0Rate an(
RiskAversion are optiona+%
1s with frontcon, ca++ing portopt whi+e specifying o&tp&t arg&ments ret&rns the
correspon(ing vectors an( arrays representing the ris., ret&rn, an( weights for each of the portfo+ios
a+ong the efficient frontier% Use them as the first three inp&t arg&ments to the f&nction porta**oc%
Ca++ing porta**oc whi+e specifying the o&tp&t arg&ments ret&rns the variance BRisk%RiskC, the
e9pecte( ret&rn BRisk%ReturnC, an( the weights BRisk%WtsC a++ocate( to the optima+ ris.y
portfo+io% )t a+so ret&rns the fraction BRisk%.ractionC of the comp+ete portfo+io a++ocate( to the
8
)f =orrowing is not (esire(, or not an option, set to *a* B(efa&+tC%
42
ris.y portfo+io, an( the stan(ar( (eviation B/vera**RiskC an( e9pecte( ret&rn B/vera**ReturnC
of the optima+ overa++ portfo+io consisting of the ris.y portfo+io an( the ris.-free asset% The overa++
portfo+io com=ines investments in the ris.-free asset an( in the ris.y portfo+io% The act&a+
proportion assigne( to each of these two investments is (etermine( =y the (egree of ris. aversion
characteriAing the investor% The va+&e of Risk%.raction e9cee(s B22WC, imp+ying that the ris.
to+erance specifie( a++ows =orrowing money to invest in the ris.y portfo+io, an( that no money wi++
=e investe( in the ris.-free asset% This =orrowe( capita+ is a((e( to the origina+ capita+ avai+a=+e for
investment% Ca++ing porta**oc witho&t specifying any o&tp&t arg&ments gives a graph (isp+aying
the critica+ points%
Si/-sto%k-in,e/ !ort"o#io
We &se the previo&s e9amp+e% 1ss&me an 1merican investor form si9-stoc.-in(e9 portfo+io an(
hePshe wants to what is hisPher optima+ portfo+io a++ocation% We ass&me ris. free investment ret&rn
is !%W, investorFs (egree of ris. aversion is ", an( =orrowing rate is 7W% We a+so a++ow short
se++ing &p to 22W of the portfo+io va+&e in each stoc. in(e9, =&t +imit the investment in any each
stoc. in(e9 to "2W of the portfo+io va+&e% *ow we &se @1T/1K to assist himPher%
First, we ca+c&+ate rate of ret&rn of the nat&ra+ +ogarithm stoc. in(e9 price, save them as a te9t
fi+e ; stoc.Zin(e9%t9t > into C5[ProgramFi+es[@1T/1K7[wor.[ stoc.Zin(e9%t9t, an( +oa( this fi+e
into @1T/1K Secon(+y, &se mean an( cov f&nctions ca+c&+ate e9pecte( ret&rn of each stoc. in(e9
an( covariance matri9 of portfo+io% Thir(+y, ca++ portcons to create the matri9 of constraints for
short se++ing a++owance an( ca++ portfopt comp&te constraine( efficient frontier% Then, Use
PortRisk, PortRet, PortWts inp&t arg&ments to the f&nction porta**oc an( fin( the optima+
ris.y portfo+io an( the optima+ a++ocation of f&n(s =etween the ris.y portfo+io an( the ris.-free
asset%
"igre X.11 M%-6%& code for asset allocation
4
c*ear a**1
c*c1
tic
2portfo*io(m
*oad stock3index(txt1
Portfo*io = stock3index (4,5!1
[m n] = si6e(Portfo*io!1
[AvrRet]= mean(Portfo*io!1
[ExpRet+ec]= ('7 AvrRet!(8')9'1
[Cov-tx]= cov(Portfo*io!1
2 C/-P:$E E..;C;EN$ .R/N$;ER ("</R$ "E,,;N A,,/WE&!
NumPorts = n1
Asset-in = [ 9' 9' 9' 9' 9' 9']1
Asset-ax = [ '(= '(= '(= '(= '(= '(=]1
pva* = '1
Constraint = portcons(>Port+a*ue>,pva*, NumPorts, >Asset,ims>, Asset-in, Asset-ax,
NumPorts!1
[PortRisk,PortRet,PortWts] = portopt(ExpRet+ec, Cov-tx, =?, [], Constraint!1
[/ptRisk,/ptRet,/ptWts]=portopt(ExpRet+ec, Cov-tx, NumPorts, [], Constraint!1
2 /P$;-A, A""E$ A,,/CA$;/N
Risk*essRate = ?(?@'1
Borro0Rate = ?(?A1
RiskAversion = =1
[Risk%Risk, Risk%Return, Risk%Wts, Risk.raction, /vera**Risk, /vera**Return] =
porta**oc (PortRisk, PortRet, PortWts, Risk*essRate, Borro0Rate, RiskAversion!
2 P,/$ $<E RE":,$
p*ot(PortRisk, PortRet, >m9>, /ptRisk, /ptRet, >x>, (((
?, Risk*essRate, >k4sBuare>, Risk%Risk, Risk%Return, >k4diamond>,(((
[?1 Risk%Risk], [Risk*essRate1 Risk%Return],>r99>!
x*aCe*(>Portfo*io Risk>!1
%*aCe*(>Portfo*io Return >!1
tit*e(>Efficient .rontier>!1
2*e#end(>Efficient .rontier>, >/ptima* Portfo*io>, >Risk free asset>, >Risk%
Portfo*io>, >Asset A**ocation Point>!
#rid on

)n this e9amp+e the c&stomer wi++ to+erate =orrowing 474%!!W of the origina+ capita+ amo&nt to
invest in the si9-stoc.-in(e9 portfo+io, an( that no money wi++ =e investe( in the ris.-free asset%
The optima+ overa++ portfo+io ret&rn is 8!%88Wwith 4?%4W stan(ar( (eviation% The e9pecte( ret&rn
of si9-stoc.-in(e9 portfo+io is 84%48W, portfo+io ris. is 8%42W% The =est a++ocation proportion of
each asset for the investor is to a++ocate #4%4!W portfo+io va+&e in SUP "22 )n(e9, !%4?W in
Swiss @ar.et )n(e9BS@)C, an( 88%?4W in Hang Seng )n(e9BHS)C, an( short se++ing 22W
portfo+io va+&e in FTS:22 )n(e9, ?!%!"W in Straits Times )n(e9BST)C ,an( #%8W in ,orea
Composite Stoc. Price )n(e9 B,JPS)C %
"igre X.12 M%-6%& otpt of optimal allocation pro0lem
48
Risk%Risk =
?(?DE?
Risk%Return =
?()EE)
Risk%Wts =
?(5EE@ 9'(???? '('@EF 9?(F@@= '())FE 9?(?5)'
Risk.raction =
@(AE@@
/vera**Risk =
?(EFE'
/vera**Return =
?(D@)D
"igre X.1( 7ptimal asset allocation
,eferences
44
Kenninga Simon, ???, Financia( Mode(ing, The @)T press, Com=ri(ge, @assach&setts, /on(on,
:ng+an( ???%
Ko(ie O%, 1% ,ane ,an( 1% V% @arc&s, 8224, Essentia(s of 3nvestment# Fifth e(ition, 99995@c'raw-
Hi++%
Ko(ie O%, 1% ,ane ,an( 1% V% @arc&s, 822", 3nvestment, Si9th e(ition, @c'raw-Hi++%
Kran(imarte Pao+o, umerica( Met4ods in Finance5 A MA0%AB+Based 3ntroduction, *ew
-or.5Vohn Wi+ey U Sons% )nc%
Cheng F% /ee, Voseph :% Finnerty, Dona+s H% Wort ,??2, Securit' Ana('sis and Portfo(io
Management, 999959999
Vac.son @ary an( @i.e Sta&nton, 8228, Advanced Mode((ing in Finance using E&ce( and
1BA,999959999%
%ppendi+ %
4!
To prove the s+ope an( the shape of e9treme corre+ation =etween two assets
1ss&ming that the weight of asset 1 an( asset K are
8
an( w w
, an(
8
w w +
%
8 8 8 8 8
8 8 8 8 8
8
8
*
8 8
@inim&m variance portfo+io B@EPC
varB C 8
B C 8 B C
% %
8

p p A B A B A B A B AB
A A A B A A A B AB
p
B AB A B
A
A A B AB A B
Min r w w w w
w w w w
F 6! w
w

+ +
+ +

+
* *

B A
w w
Case one5 Two assets are perfect+y position corre+ate(
)f short sa+es are a++owe(, then even tho&gh

Se++ing short asset 1 an( go e9tra +ong in asset K
8
)f , +
8 8
8
*
B C B C B C B C
L B C M
L B C M
5 2
B C
, B C B C
B C
B C
B C B C
p A A A B
p A A A B
p A A A B
p
B
A
A A B
p p
A B A B
A A
p
p
A A B
p
p A B
A
E R w E r w E r
w w
w w
M1P w
w
E R
E r E r
w w
E R
E R
w E r E r
w

+
+
+

Case Two5 Two assets are perfect+y negative corre+ate(

4"
V&st invest =oth asset 1 an( = B=oth in +ong positionC
8
)f ,
8 8
8
*
* 8
*
* *
B C B C B C B C
L B C M
L B C M
5 2
a% 2
=%
8
8
8
B C 2
L B C M B
p A A A B
p A A A B
p A A A B
p
B
A
A A B
A A p
A A
B
A A A
A B
B
p A B B B
A B
p A A A B A
E R w E r w E r
w w
w w
M1P w
w
w w
w w
%et w w w
w w w

+

t

+

>
>
+
+ >
+
+
*
*
* *
C

L B C M
B C
B C
B C B C

c%
2%"
2%"
2%"
B C 2%" 2
L B
A B B
A A
p A A A B
p
p p
A A B
A B
p
A p B B
A
A A
B
A A A
A B
B
p A B B B
A B
p A A A
w4en w w
w w
E R
E R
w E r E r
w
w
w w
%et w w w
w w

+
>
+

+
+ +

+ +

<
<
+
+ <
+

*
C M L B C M

L M
B C
B C
B C B C

B C
B A A B B
A A
p A A B A B
p
p p
A A B
A B
p
A p A B
A
w
w4en w w
w w
E R
E R
w E r E r
w
w

+
<
+

+
+

%ppendi+ &
4#
The weights of minim&m variance portfo+io
The weights sho&+( =e chosen so that Bfor e9amp+eC the ris. is minimiAe(, that is
8 8 8 8 8
@in 8
A
P A A B B A B AB A B
w
w w w w + +
for each chosen ret&rn an( s&=Gect to
, 2, 2
A B A B
w w w w +
% The +ast two constraints simp+y
imp+y that the assets cannot =e in short positions%
S&=stit&te
B , C
A B
AB
A B
!ov r r

into eD&ation I%#%

8 8 8 8 8
B C 8
p A A A B A B A B AB
w w w w + +
8 8 8 8 8 8 8 8
8 8 8
p A A B A B A B A AB A B A B AB A B
w w w w w w + + +
F%J%C%
8
2
P
A
w

8
8 8 8
8 8 8 8 !
P
A A B A B AB A B A AB A A
A
w w w
w

+ +

H
8 8 8
8 B 8 C 8 8 2
A A B AB A B B AB B A
w +
8
8 8
8
B AB A B
A
A B AB A B
w

B A
w w
When QH

B
A
A B
w

8
A B
B
A B
w

When QH -

B
A
A B
w

+

A
B
A B
w

+

When QH 2

8
8 8
B
A B
A
w

+

8 8
8 8
8
A B
A B
B
w

+

47