He, G. y Litterman, R.

(
2002 ) . The intuition behind Black - Litterman
model portfolios (
27 p.) [
s. l] : [ s. e.] (
Working PapersSeries, ). (
040949)
THEINTUITIONBEHINDBLACK-LITTERMANMODEL
PORTFOLIOS
by
Guangliang He
Goldman SachsAsset Management
32 OldSlip, 24th Floor
NewYork, NY10005
guangliang. he@gs. com
Robert Litterman
Goldman SachsAsset Management
32 Old Slip, 24th Floor
NewYork, NY10005
bob. litterman@gs. com
Abstract
In thisarticle we demonstrate that the optimal portfoliosgenerated by the Black-Litterman asset
allocation model have a very simple, intuitive property. The unconstrained optimal portfolio in the
Black-Litterman model isthe scaled market equilibrium portfolio( reflecting the uncertainty in the
equilibrium expected returns) plusa weighted sum of portfoliosrepresenting the investor' s views.
The weight on a portfoliorepresenting a viewispositive when the viewismore bullishthan the one
impliedby the equilibrium andthe other views. The weight increasesasthe investor becomesmore
bullish on the view, and the magnitude of the weight also increasesasthe investor becomesmore
confident about the view.
1 Introduction
Since publication in 1990, the Black-Litterman asset allocation model hasgainedwide application in
many financial institutions. Asdevelopedin the original paper, the Black-Litterman model provides
the flexibility of combining the market equilibrium with additional market viewsof the investor.
The Black-Litterman approachmay be contrastedwiththe standardmean-variance optimization in
whichthe user inputsa complete set ofexpected returns' andthe portfoliooptimizer generatesthe
optimal portfolio weights. Because there isa complex mapping between expected returnsand the
portfolioweights, andbecause there isnonatural starting point for the expectedreturn assumptions,
usersofthe standardportfoliooptimizersoften findtheir specification ofexpectedreturnsproduces
output portfolio weightswhich do not seem to make sense. In the Black-Litterman model the user
inputsany number ofviews, whichare statementsabout the expectedreturnsofarbitrary portfolios,
and the model combinesthe viewswithequilibrium, producing boththe set ofexpected returnsof
assetsaswell asthe optimal portfolioweights.
AlthoughBlack and Litterman concluded in their 1992 article [ Black and Litterman, 1992] :
our approach allowsusto generate optimal portfoliosthat start at a set ofneutral
weightsand then tilt in the direction ofthe investor' sviews."
they didnot discussthe precise nature ofthat phenomenon. Aswe demonstrate here, the optimal
portfolio for an unconstrained investor isproportional to the market equilibrium portfolio plusa
weighted sum of portfoliosreflecting the investor' s views. 2 Now the economic intuition becomes
very clear. The investor startsby holding the scaled market equilibrium portfolio, reflecting her un-
certainty on the equilibrium, then investsin portfoliosrepresenting her views. The Black-Litterman
For simplicity, we use ` return' torefer to`excessreturn over the one period risk free rate'.
2Please refer to Section 2 for precise definition ofa view.
1
model computesthe weight toput on the portfoliorepresenting each viewaccording tothe strength
ofthe view, the covariance between the viewand the equilibrium, and the covariancesamong the
views. We showthe conditionsfor the weight on a viewportfolio tobe positive, negative, or zero.
We also show that the weight on a view increaseswhen the investor becomesmore bullish on the
view, andthe magnitude ofthe weight increaseswhen the investor becomeslessuncertain about the
view.
The rest of the article isorganized asfollows. In Section 2, we reviewthe basicsofthe Black-
Litterman asset allocation model. Then we present our main resultsin Section 3. Numeric examples
are presented in Section 4 to illustrate our results. Finally, we conclude the article in Section 5.
2 The Black-Litterman Asset Allocation Model
The Black-Litterman asset allocation model usesthe Bayesian approachtoinfer the assets' expected
returns[ Black and Litterman, 1990, 1992] . Withthe Bayesian approach, the expected returnsare
random variablesthemselves. They are not observable. One can only infer their probability distribu-
tion. The inference startswitha prior belief. Additional information isused along withthe prior to
infer the posterior distribution. In the Black-Litterman model, the CAPMequilibrium distribution
isthe prior, the investor' s viewsare the additional information. The Bayesian approach isused
to infer the probability distribution of the expected returnsusing both the CAPMprior and the
additional views.
Let' sassume that there are N-assetsin the market, whichmay include equities, bonds, currencies,
and other assets. The returnsofthese assetshave a normal distribution with
being the expected
2
return and Ethe covariance matrix'. That is
1) r N(, E),
where r isthe vector of asset returns. In equilibrium, all investorsas a whole hold the market
portfolio W, Q . The equilibrium risk premiums, II, are such that ifall investorshold the same view,
the demand for these assetsexactly equalsto the outstanding supply[ Black, 1989] . Assuming the
average risktolerance ofthe world isrepresented by the risk aversion parameter 8, the equilibrium
risk premiumsare given by
2) 11=
SEweq.
The Bayesian prior isthat the expected returns, , are centered at the equilibrium values, that is
they are normally distributedwiththe mean ofII,
3)
p=
11+ e( e),
where E( e) isa normally distributedrandom vector withzeromean and covariance matrix TE, where
T isa scalar indicating the uncertainty ofthe CAPMprior.
In addition tothe CAPMprior, the investor alsohasa number ofviewson the market returns. A
viewisexpressed asa statement that the expected return ofa portfoliophasa normal distribution
withmean equal toq anda standard deviation given by w. 4 Let Kbe the total number ofthe views,
P be a Kx Nmatrix whose rowsare these portfolioweightsand Q be a K-vector ofthe expected
returnson these portfolios. That is
4)
P1 _ (
pi p2 ... pK)
5)
Q / (
ql q2 ... qK)
3The covariance matrix Eisassumed tobe a known constant matrix.
4We also refer to the portfoliopasa viewportfolio.
3
Then the investor' sviewscan be expressed as
6) P=
Q +
6( V ),
where EM isan unobservable normally distributed random vector with zero mean and a diagonal
covariance matrix Q .' It isfurther assumed that the E( e) and EMare independent
TE 0
7) N 0,
E( V ) ) 0 Q
These viewsthen are combinedwiththe CAPMprior in the Bayesian framework. The result isthat
the expected returnsare distributed asN(, M- 1), where the mean
isgiven by
8) TE)-
1 +
Pl52- 1P]
1 [(
TE)- 1H+
PIQ - 1Q ] ,
and the covariance matrix M- 1 isgiven by6
1
9) M- 1 =
L(
TE)-
1 +
P'-
1PJ
3 The Unconstrained Optimal Portfolio
In practice, it isquite straight forward to apply equation ( 8) for calculating the mean ofexpected
returns. However, it isoften difficult to findthe original economic intuitionsofthe viewsfrom these
The assumption ofa diagonal Q matrix isnot a restriction. When the matrix Q isnot diagonal, it can alwaysbe
decomposedintothe form Q = V S2V ` where! 2 isdiagonal. The original viewscan be transformedintoP=
Q +&)
where P= V - 1P, Q = V - 1Q , and the covariance matrix ofElul is! 2 which isdiagonal. In fact, the assumption of
diagonal Q matrix isnot needed throughout thisarticle. Only in the proofofProperty 3. 1, the Q matrix needstobe
block diagonal. See the prooffor the details.
6Aspointedout by S. Satchell andAScowcroft[ Satchell and Scowcroft, 1997] , equation( 9) isa well known result,
despite that it isnot displayedin Black and Litterman' soriginal papers.
4
numbers, particularly when the number ofassetsare large. The trouble withtrying to make sense
out ofthe expectedreturnsofdifferent asset classesisthat they are all relatedtoeach other through
their relative volatilitiesand correlations. For example, it iseasy to conjecture that the German
equity market will outperform the rest ofEuropean markets, but muchlesseasy toimagine what the
implication ofthat conjecture isfor the relative expected returnsofthe German andother markets.
To increase the expected return on German equitiesand hold all other expected returnsfixed does
not, in the mathematicsof an optimizer, suggest an overweight in German equitiesrather, it
suggeststo the optimizer that by using the relative volatilitiesand correlationsof the different
marketsit can create a muchmore complicated portfoliowithhigher expectedreturn and lower risk
than would be available simply by creating an overweight of the German market.
In the Black-
Litterman approach a view that German equitieswill outperform the rest ofEuropean equitiesis
expressedasan expectation ofa positive return on a portfolioconsisting ofa long position in German
equitiesandmarket capitalization weightedshort positionsin the rest ofthe European markets. This
viewisthen translatedthroughequation ( 8), which appropriately takesvolatilitiesand correlations
into account, into adjustmentsto expected returnsin all ofthe markets. Aswe showbelow, these
adjustmentsto expected returnsare exactly those needed tosuggest tothe optimizer that the best
opportunity isa simple overweighting ofthe German equity market, financedby underweighting the
rest ofEuropean markets. These typesofcomplex transformations, from viewson portfoliosto the
expected return vector, and from the expected return vector to the optimal portfolio, are generally
difficult to understand. Thus, instead oflooking at the expected returnsdirectly, we examine the
Black-Litterman optimal portfolio weights. We begin with the case of an unconstrained investor
withrepresentative risk aversion parameter equal tob. Thisportfoliogivesussome very interesting
insightsabout the Black-Litterman asset allocation model.
5
Because the expected returnsare random variablesthemselvesin the Black-Litterman model,
the distribution ofthe returnsisno longer simply N(,
E).
According toequation ( 1), (
8), and( 9),
the distribution for the returnsis
10) r - N(, E),
where E= E+ M- 1.
Given the mean
and the covariance matrix 2, the optimal portfolio can be constructed using
the standardmean-variance optimization method. For an investor withthe risk aversion parameter
S, the maximization problem can be written as
11) max w'- 6w' Ew.
The first order condition yieldsthat
12) bow*
or equivalently,
13) w*
S
where w* isthe vector ofthe optimal portfolio weights. Using equation ( 8),
the optimal portfolio
weightscan be written as
14) w*_
3
E- 1M- 1 [( TE)-
1+
P' Q - 1 Q
Note the fact that
15) E- 1 = ( E+ R-1)-
1 =
M- M( M+ E- 1)-
1
M
the term - 1M- 1 can be simplified as
16) E 1M 1
1+ T
CI- P' A-
1P1+ T)
6
where matrix A= Q lr+ PE/( 1 + T) P'. The optimal portfolioweights, w*= i-1/ 6 now can be
written as
17) w*
1 + T(
weq +
P' x A)
where weq = (
SE)- 1II isthe market equilibrium portfolio and Aisa vector defined as
18) A=
TQ - 1Q / 8-
A- 1P
1+ T
weq -
A- 1P
1 + T
P' TQ - 1Q 16.
Since each column ofmatrix P isa viewportfolio, equation ( 17) showsthat the investor' soptimal
portfolio isthe market equilibrium portfolio weq, plusa weighted sum ofthe portfoliosforming the
views, then scaledby a factor of1/(
1+ T).
The weight for eachportfolioisgiven by the corresponding
element in the vector A.
There are intuitive interpretationsofthe weightsin equation ( 18). The first term showsthat the
stronger the viewis( either with a higher expected return qk, or a lower level ofuncertainty, i. e. a
higher value ofthe precision, Wk/ T),
the more weight it carriesin the final optimal portfolio. The
secondterm showsthat the weight ofa viewispenalizedfor the covariance between the viewportfolio
and market equilibrium portfolio. Since the market equilibrium information isalready presented by
the prior, the covariance ofa viewportfoliowiththe market equilibrium portfolioindicatesthat the
viewcarrieslessnewinformation and the penalty makessense. Similarly, the last term showsthat
the weight ispenalized for the covariance ofa viewportfoliowithother viewportfolios. Again, since
the covariance between these viewportfoliosindicatesthat the information isin a sense being double
counted, it isintuitive that there should be a penalty on the weight associated with an increased
covariance withother viewportfolios.
For an investor witha different risktolerance, the optimal portfoliow* can be obtainedby scaling
7
the portfolio w*
where S isthe risk aversion parameter for the investor. For an investor with a fixed risk( standard
deviation) limit v, the portfolio optimization isformulated as
20) max w'.
W/ rW<a2
The optimal portfoliow* can also be obtained by scaling the portfolio w*
21) w*= (
o-61
ww
p) w*.
For an investor with other constraintson the portfolio, the optimal portfolio can be obtained by
using the usual portfolio optimization package with
and 2 asinputs.
Since the weight Aon a viewplaysa very important role in the portfolioconstruction process, we
needtoknowwhen the weight ispositive, and howthe weight changes. The following twoproperties
showjust what we wanted.
Property 3. 1 Let P, Q , and Q represent the Kviewsheld by the investor initially,
be the mean
ofexpectedreturnsby using these viewsin the Black-Litterman model, Abe the weight vector defined
by equation ( I8). Assume the investor nowhasone additional view, represented by p, q, andW. For
the case ofK+ 1 views, the new weight vector isgiven by
22)
A
K+ lAlb
=
K+ 1
where b= P
11+ Tp, c=
w17-+
p' 1 Ep, and K+ 1, the weight on the additional view is
23) K+ 1 = (
cb' Aib)b.
8
Let
24)
since w*= ( bE)
l
isthe unconstrained optimal portfolio ofthe Black-Litterman model with the
first Kviews, =
SEw* can be seen asthe implied expected returnsofthe first Kviews, even though
usually isdifferent from , due to the difference between E and E. In the limit asT and Q go to
zero while Q 1,F staysfinite, A and
become identical.
Since cb' Alb>
0, equation ( 23) showsthat AK+ 1i the weight on the additional viewwill have
the same sign asthe expression q p'. Thismeansthat the weight K+ l ispositive ( negative)
when the strength ofthe view on the portfoliopismore bullish( bearish) than the portfolio' simplied
expected return p'. The additional viewwill have a zero weight ifq= p'.
In thiscase, the weights
on the first Kviewsare identical to the weightsgenerated in the Black-Litterman model with the
first Kviews.
Proof. See Appendix.
Property 3. 2 For a particular viewk, itsweight Ak isan increasing function ofitsexpected return
qk. The absolute value ofAk isan increasing function ofitsconfidence level Wk
1
Proof: See Appendix.
4 Numeric Examples
In thissection, we present several numeric examplesto illustrate our results. In all examplesthe
market consistsofthe equity indexesofseven major industrial countries. 7 The correlationsbetween
the index returnsare listed in Table 1, and the index return volatilities, the market capitalization
7We ignore the bondmarkets, currency markets, and other equity marketsfor simplicity.
9
weights, and equilibrium risk premiumsare listed in Table 2.
The S parameter, representing the
world average risk aversion isassumed to be 2. 5.
First, let usexamine the example ofa viewon the German equity market presented in Section 3
withthe traditional mean-variance approach. Suppose the investor isnot familiar withthe Black-
Litterman model. Nevertheless, let usassume that she usesthe CAPMasher starting point. When
she doesnot have other views, she assumesthat the expected returnsequal the equilibrium risk
premiumsas shown in Table 2. Asa result, she holdsthe market portfolio. With a view that
the German equity market will outperform the rest ofEuropean marketsby 5%
a year, she needs
to modify the expected returns.
She might proceed asfollows.
To be precise in expressing her
view, she setsthe expectedreturn for Germany 5%
higher than the ( market capitalization) weighted
average ofthe expectedreturnsofFrance and the UnitedKingdom. She keepsthe market weighted
average expected returnsfor the European countriesandthe spread between France andthe United
Kingdom unchanged from their equilibrium values. She also keepsthe expected returnsfor non-
European countriesunchanged from their equilibrium values. Since the equilibrium already implies
that Germany will outperform the rest ofEurope, the changesin expected returnsare actually quite
small, asshown in Table 3. However, the optimal portfolio isquite different from what one would
have expected: although the increasedweight in the German market and the decreased weightsin the
United Kingdom and France marketsare expected, the size ofthe changesare quite dramatic. Even
more puzzling are the reduced weightsin Australia and Canada aswell asthe increased weightsin
J apan and the United States. Since the investor doesnot have any view on these countries, why
should she adjust the weight in these countries? Presumably it isbecause ofthe way the viewsare
being translatedinto expected returns. Aswe can see, the investor hasalready taken extra care in
her approach totranslating the viewintoexpected returns. Can she possibly do any better?
10
Withthe Black-Litterman approach' sthe same viewisexpressedasan expectation ofa 5% return
on a portfolio consisting of a long position in German equity and market capitalization weighted
short position in rest ofthe European markets. The Black-Litterman expectedreturnsare calculated
by equation ( 8), and the optimal portfolio weightsare calculated by ( SE)
l.
The viewportfolio,
the expected returns, the optimal portfolio, and the deviation from her initial portfolioweightsare
shown in Table 4. The optimal weight vector isvery intuitive in thiscase. Being a Bayesian, the
investor hasstarted with a scaled market portfolio we9l( 1 +
T),
reflecting her uncertainty on the
CAPM. 9 Since she expressed a view on the portfolio of Germany versusthe rest of Europe, she
simply addsan exposure tothe viewportfolioin addition toher initial portfolio, the scaled market
equilibrium portfolio. The size ofthe exposure isgiven by A/( 1+
T).
The weight Aiscalculated by
the Black-Litterman model with equation ( 18). In thisexample, the value of A is0. 302. One can
easily verify in the Table 4, the optimal deviation isexactly the viewportfolio multipliedby A.
Now, let' sassume that the investor hasone more view. In additional tothe first viewofGermany
versusthe rest ofEurope, she also believesthat the Canadian equitieswill outperform the US equity
by 3%
a year. Thisviewisexpressed asan expected 3%
annualized returnsfor the portfoliooflong
Canadian equitiesand short US equities. Table 5 showsthe views, the expected returns, and the
optimal portfolio. Again, the resultsclearly showthat the optimal deviation isa weighted sum of
the viewportfolios. The corresponding weightsAl/(
1+ T),
A2/(
1+ T) are equal to0. 284 and 0. 398,
respectively. Again, the result isvery easy to understand, the investor hastwo views, so she adds
exposuresto these portfolios. The Black-Litterman model givesthe optimal weights.
Assume nowthat the investor becomesmore bullish on the Canada/ US view. Instead of the
BThe parameter T isassumed to be 0. 05, whichcorrespondstothe confidence level ofthe CAPMprior mean ifit
wasestimated using 20 yearsofdata.
9In the case ofno additional views, the quantities and 2 are reduced to11 and( 1+ 7-) E.
11
3%
outperformance, she believesthat the Canadian equitieswill outperform the US market by 4%.
According to Property 3. 2, the weight A2 on the second portfolio should increase.
Asshown in
Table 6, the value ofA2 increased from the value of0. 418 in the Table 5 to0. 538.
Assume the investor now becomeslessconfident on the view of Germany versusthe rest of
Europe. She isonly 50% asconfident in the viewnow. The isreflected by a change ofcall from
0. 021 to0. 043. Her viewon Canada/ US isunchangedat 4%.
According toProperty 3. 2, the absolute
value of weight Al on the first portfolio should decrease. Table 7 showsthe value decreased from
0. 292 to0. 193.
Our last example illustratesProperty 3. 1. In addition tothe two viewsin Table 7, the investor
hasa thirdview. The thirdviewisthat the Canadian equitieswill outperform the J apanese equities
by 4. 12%. But following equation ( 24), the implied expectedreturn ofthe portfolio oflong Canada
and short J apan isexactly 4. 12%
when the investor had two views. The expected returnsofthird
viewportfolio matchesexactly the expected returnsimplied by the Black-Litterman model using
only twoviews. According toProperty 3. 1, the weightson the first twoviewsshouldnot be changed
and the weight on the thirdviewshouldbe zero. Table 8 doesshowthat the weight A3 on the third
viewiszero andthe expected returns, optimal portfolioweightsare exactly the same asin Table 7.
5 Conclusion
Instead of treating the Black-Litterman asset allocation model asa rocket science black box which
generatesexpected returnsin some mysteriousway, we presented a method to understand the
intuition ofthe model. Withthe newmethod, investing using the Black-Litterman model becomes
very intuitive. The investor wouldinvest in the scaled market portfoliofirst, then addthe portfolios
12
representing the views.
The Black-Litterman model givesthe optimal weightsfor the portfolios.
When there are constraints, or the risktolerance isdifferent from the world average, one can always
use the expectedreturns, andthe covariance matrix Ein a portfoliooptimization package toobtain
the optimal portfolio. Now, the mean ofexpected returns
issimply the matrix product of Eand
the optimal portfolio w*= (
weq+
P' A)/( 1 + ,
r) scaled by the average risk aversion parameter S.
Unlike a standardmean-variance optimization, the Black-Litterman model, ifproperly implemented,
will alwaysgenerate an optimal portfolio whose weightsare relatively easy to understand.
13
Appendix
ProofofProperty 3. 1
In the case where the investor holdthe initial Kviews, the weight vector Acan be written as
25) A=
TS2- 1Q / 6-
A- 1P
1 + T(
weq +
PITS2- 1Q 6) .
Similarly, in the case where the investor hold an additional view, the weight vector is
26) A-
1P1+ T(
weq +
P' TSZ - 1Q 1S) .
where P, Q , S2 and Aare the counterpart ofP, Q , Q , and Ain the case ofK+ 1 views. They can
be written as10
P
27)
P =
P/
Q
28)
Q =
q
S2 0
29)
0 w
and
A b
30) A=
V c
Using equation ( 8),
after a fewstepsofalgebra, equation ( 26) can be written as
1
31)
TQ - 1Q 1A-
1 P+ PEP' TQ - 1P+
PEpw- lq
S
Tw_1q
1+ 7
p, p+ p' EP' TQ - 1Pp+ p' Epw- 1q
1OThe assumption of diagonal matrix S2 isnot needed here. It only needsit to be block diagonal so equation ( 29)
holds.
14
The inverse ofAis
32) - 1
A- 1 + A- 1bb' A- 1/ d - A- 1b/ d
b' A- 1/ d 11d
where d= ( c- b' A- 1b).
Applying the last row ofA- 1 in equation ( 31) givesthe last element ofA
as
TW- 1 q (
b' A- 1 - 1 )
P+ PEP' TQ - 1P+
PEpw- 1q
33)
K+ 1 =
b + ( 1+ T) dS
p'+ p' EP' TQ - 1Pp+ p' Epw- 1q
After expanding the matrix multiplication and a few stepsof algebra, the above equation directly
leadsto
E-
1-
34)
K{1 (
c- b' A- 1b) 6'
which isexactly equation ( 23). Similarly, applying the first KrowsofA- 1 in equation ( 31) gives
the first KelementsofAas
A =
TQ _1Q I A- 1 +
A- 1bb' A- 1 Alb
J
P
cb' A- 1b cb' A- 1b/
p'
35) X
1 + T(
6Weq +
P' TQ - 1Q + pTC.w- 1q) .
After some simple but lengthy algebra, Acan be simplified intothe following form,
36) t1 = A-
K+ 1A- 1b
Combining equation ( 34) and equation ( 36), it iseasy tosee that
37)
A-
K+ 1A- 1b
=
K+ 1
Since matrix S2 ispositive definite and PEP' ispositive semi-definite, the sum ofthe two, matrix A
ispositive definite, so isA- 1. Andsince c- b' A- 1b isa diagonal element ofA- 1, it must be positive.
15
Therefore, K+ 1, the weight on the last viewhasthe same sign asthe numerator in equation ( 34).
In the limit when T and 0go to zero and! Z / T stay constant, the numerator becomesqp'.
ProofofProperty 3. 2
Notice that equation ( 18) can be rearranged as
38) A=
A- 1Q 16
A- 1P
l+
Tweq-
It isclear that
39)
aaz _
1 (
A 1)
aqk
ik
in particular
40)
aak_
1 (
A- 1)
aqk
6
kk
Because ( A- 1) kk isa diagonal element ofa positive definite matrix, it must be positive. Therefore,
Ak isa increasing function ofqk.
Using equation ( 38),
the partial derivative aA/ awk1 can be written as
aA _aA E
41)
Ow-
aWk
1
CQ /
SP
1+ T
we q)
Because aA- 1 10w- 1 = A- 1( aA1aw- 1) A- 1 and aA/ awkl =
WkIkk/ T, equation ( 41) leadsto
42)
al = (
Wk/ T) A- 16kkA.
k
where tkk isa matrix with1 asthe k-thdiagonal element and zeroeverywhere else. The k-thelement
ofthe above partial derivative is
aA
43)
aW
kl (
Wk/ T) ( A-
1)
kk
Ak-
It hasthe same sign asAk. For a positive ( negative) Ak, increasing Wk1 wouldcause Ak to increase
decrease).
In other words, the absolute value ofAk isan increasing function ofwkl
16
References
Fischer Black and Robert Litterman. Asset allocation:
Combining investor viewswith market
equilibrium. Fixed income research, Goldman Sachs, September 1990.
Fischer Black and Robert Litterman. Global portfolio optimization. Financial AnalystsJ ournal,
pages28- 43, September-October 1992.
Fisher Black. Universal hedging: Optimizing currency risk andreward in international equity port-
folios. Financial AnalystsJ ournal, pages16- 22, J uly-August 1989. missing.
S. Satchell and AScowcroft. Ademystification ofthe black-litterman model; managing quantitative
and traditional portfolio construction. Research papersin management studies, J udge Institute
ofManagement Studies, University ofCambridge, September 1997.
17
Table 1: Correlationsamong the equity market index returnsofthe seven countries.
Australia Canada France
Germany
J apan UK
Canada 0. 488
France 0. 478 0. 664
Germany
0. 515 0. 655 0. 861
J apan 0. 439 0. 310 0. 355 0. 354
UK 0. 512 0. 608 0. 783 0. 777 0. 405
USA 0. 491 0. 779 0. 668 0. 653 0. 306 0. 652
18
Table 2: Annualized volatilities, market capitalization weights, and equilibrium risk premiumsfor
the equity markets.
a
we9
II
Australia 16. 0 1. 6 3. 9
Canada 20. 3 2. 2 6. 9
France 24. 8 5. 2 8. 4
Germany
27. 1 5. 5 9. 0
J apan 21. 0 11. 6 4. 3
UK 20. 0 12. 4 6. 8
USA 18. 7 61. 5 7. 6
19
Table 3: The expectedreturns, optimal portfolioweights, and their changesfrom equilibrium values
in the traditional mean-variance approach. The expected returnsare adjusted to reflect the view
that Germany equitieswill outperform the rest ofEuropean equitiesby 5%.
WoPt
H
woPt wee
Australia 3. 9 - 5. 1 0. 0 6. 7
Canada 6. 9 - 2. 3 0. 0 4. 5
France 7. 6 - 50. 1 - 0. 8 55. 3
Germany
11. 5 83. 6 2. 4 78. 1
J apan 4. 3 14. 9 0. 0 3. 3
UK 6. 0 - 22. 8 - 0. 8 35. 2
USA 7. 6 66. 6 0. 0 5. 1
20
Table 4: The first column isthe portfolio used by the investor to expressthe view on Germany
versusthe rest ofEurope in the Black-Litterman approach. The expected return ofthisportfolio is
5%.
The relative uncertainty ofthe viewisrepresented by the value ofw/ T, whichequals0. 021. The
rest oftable showsthe expected returnscalculated by equation ( 8),
the optimal portfolio( SE)'
p,
and the deviation ofthe portfolio from the initial portfoliowegl( 1 +
T).
1+ T
Australia 0. 0 4. 3 1. 5 0. 0
Canada 0. 0 7. 6 2. 1 0. 0
France - 29. 5 9. 3 - 4. 0 - 8. 9
Germany
100. 0 11. 0 35. 4 30. 2
J apan 0. 0 4. 5 11. 0 0. 0
UK - 70. 5 7. 0 - 9. 5 - 21. 3
USA 0. 0 8. 1 58. 6 0, 0
21
Table 5: The first two columnsare the twoviewsheld by the investor. The rest oftable showsthe
expectedreturnscalculatedby equation ( 8), the optimal portfolio( SE)-
l,
andthe deviation ofthe
optimal portfolio from the scaled equilibrium weights.
P/ W. w*_
Weq
1+,
Australia 0. 0 0. 0 4. 4 1. 5 0. 0
Canada 0. 0 100. 0 8. 7 41. 9 39. 8
France - 29. 5 0. 0 9. 5 - 3. 4 - 8. 4
Germany
100. 0 0. 0 11. 2 33. 6 28. 4
J apan 0. 0 0. 0 4. 6 11. 0 0. 0
UK - 70. 5 0. 0 7. 0 - 8. 2 - 20. 0
USA 0. 0 - 100. 0 7. 5 18. 8 - 39. 8
q 5. 00 3. 00
w/' r 0. 021 0. 017
A 0. 298 0. 418
22
Table 6: The Canada/ US viewismore bullishthan the previouscase. The expected returnson the
viewis4% instead of3% shown in Table 5.
P/ w` w*-
wev
1+ T
Australia 0. 0 0. 0 4. 4 1. 5 0. 0
Canada 0. 0 100. 0 9. 1 53. 3 51. 3
Rance - 29. 5 0. 0 9. 5 - 3. 3 - 8. 2
Germany
100. 0 0. 0 11. 3 33. 1 27. 8
J apan 0. 0 0. 0 4. 6 11. 0 0. 0
UK - 70. 5 0. 0 7. 0 - 7. 8 - 19. 6
USA 0. 0 - 100. 0 7. 3 7. 3 - 51. 3
q
5. 00 4. 00
w/ T 0. 021 0. 017
A 0. 292 0. 538
23
Table 7: In thiscase, the investor becomeslesscertain about the view on the Germany versusthe
rest ofthe Europe. It isrepresented by the value ofw/ T being double from the value in Table 6.
P/ 2U w*-
W= q
1+'
Australia 0. 0 0. 0 4. 3 1. 5 0. 0
Canada 0. 0 100. 0 8. 9 53. 9 51. 8
France - 29. 5 0. 0 9. 3 - 0. 5 - 5. 4
Germany
100. 0 0. 0 10. 6 23. 6 18. 4
J apan 0. 0 0. 0 4. 6 11. 0 0. 0
UK - 70. 5 0. 0 6. 9 - 1. 1 - 13. 0
USA 0. 0 - 100. 0 7. 2 6. 8 - 51. 8
q 5. 00 4. 00
w/ T 0. 043 0. 017
A 0. 193 0. 544
24
Table 8: In thiscase, the investor hasthree views. However, since the thirdviewisalready implied
by the equilibrium and the first twoviewsasshown in Table 7, the weightson the first twoviewsare
the same asthe previouscase. The optimal portfolio weightsin thiscase isidentical to the values
ofthe previouscase.
P/ w* w*_
wev
1+ T
Australia 0. 0 0. 0 0. 0 4. 3 1. 5 0. 0
Canada 0. 0 100. 0 100. 0 8. 9 53. 9 51. 8
Rance - 29. 5 0. 0 0. 0 9. 3 - 0. 5 - 5. 4
Germany
100. 0 0. 0 0. 0 10. 6 23. 6 18. 4
J apan 0. 0 0. 0 - 100. 0 4. 6 11. 0 0. 0
UK - 70. 5 0. 0 0. 0 6. 9 - 1. 1 - 13. 0
USA 0. 0 - 100. 0 0. 0 7. 2 6. 8 - 51. 8
q
5. 00 4. 00 4. 12
w/' r 0. 043 0. 017 0. 059
A 0. 193 0. 544 0. 000
25