MASSACHUSETTS

INSTITUTE OF TECHNOLOGY

DEPARTMENT OF ECONOMICS
'O MEMORIAL DRIVT
C A M B R I D G E , M A S S A C H U S E T T S0 2 1 4 2 . 1 3 4 7

12May 2008

ProfessorWilliam Ziemba
University of British Columbia
SauderSchoolof Business
2053Main Mall
Vancouver, BC
CanadaV6T LZz

Dear Bill,

I marvel that you still cherish hankerings for the Kelly-Breiman-Latand criterion.

I enclosea 2005letter from me to Elwyn Berlekamp of Berkeley, hoping to set him

straighton that topic. Also, I have marked up by ballpoint pen your one-pageemail
letter to me.

I don't perceive that in 1986 you made any case for persons who don't have
logarithmic utility to respectthe injunction that they should act as if they di,l have
it. My 2005 letter spells out the metric harm that possessorsoI 4W or A-L/W
utilities would do to themselvesif they listenedto Kelly or Jim Thorpe or ... .

Best,

Paul A. Samuelson

Encs.(2)
 M A SS A C H U S E T T SI N S T I T U T E O F T E C H N O L O G Y

D E P A R T M E N TO F E C O N O M I C S
50 MEMORIAL DRIVE
0 2 14 2- 13 4 7
C A M B R I D G E ,M A S S A C H U S E T T S

16 November2005

ProfessorElwyn Berlekamp
2039 ShattuckAvenue,#408
Berkeley,CA 94704

Dear Elwyn,

I'm gratefulfor the reprintof your recentfne AmericanScientislbook review. Exceptfor your
thoughtfulness,I'd have likely not learnedabout it.

Here is a fablethatmight be of someinterest.I havethree"rational"neighbors:Tom, Dick and

Harriet. All are risk averterswho shunall "unfair" gambles.Being of unequaldegreeof risk
aversion,Harriet is the most cautious;Dick (Goldilocks)hasmore risk tolerancethan Harriet
but lessrisk tolerancethan sportiveTom. All face one seriousproblem. They must invest for
a future period of retirementwhile the feasiblechoiceis between(1) safecashwith only zero
yield, and(Z) one singlestockthat in everyperiodwill, for each$1 investedin it now, bring
to the investorone period from now either $4 or $i " even odds.

Test I. I testmy neighborin variousways. "If you mustput L00%of your nesteggin only one
of thesetwo options,which will you pick?" Tom replies:1,00%in the stock,x : L; Harriet
replies:100%in safecash,L-x : 1. Dick says:I'm indffirent betweenali in the volatilestock
and ail in the safecash.

Q. Given a horizon of N ) 1 period.suntil the final dateof your retirement,will you change
your x proportions?To the surpriseof lay people, all three answerNo, no change.

Test II. Now you're giventhe optionof blendingcashand stock.Harriet replies:for everyN,
large or small, I'11put ?* = x* in the stockand Z* = 1-x* in safecash.No surprisewhen
Tom reportsa larger x" than Harriet: for Tom, x* : 1* and 1-x* : 0*. Middling Dick, as
expected,hashis x* betweenthoseof the othertwo. For Dick, ** : ** : 1-x*.(MaybeDick's
last nameis Kelly or Breimanor Latan6.)

Why theseparticulardecisions?Interrogationsrevealthat all three are devoutLaplacians,who

ever strive toward maximal ExoectedConcaveUtilitv of wealth outcomes.

vtax{lu(+l.*u{l)}
z
in TestI, U/ > 0 > U// ( 1 .1 )
* f/ \+l)
Mr. Elwyn Berlekamp
Page2
16 November2005

u"x{jut+x+1-x).}u(}*.t -*)}, in restII. (r.z)

Dick reportshis U(W) : log W (what economistscall L738D. Bernoulliutilify). What's the
samething, Dick is a GeometricMean maxim:r,er.

CautiousHarrietis a HarmonicMeanmaximizer,with U = -W-1*A. Her limiteddegreeof risk

tolerancefits pretty well lots of empiricalWall Street"equity premia" data.

To measure how mucheachpersongainsor losesin (subjective)"CertaintyEquivalentDollars,"

I definetheir threedifferent CE's by the following generalformulas:

: iuto).iu(i)
u(cE) (2.1)
cE : u-'f1u(+).iu/r\l
2
= E(4,1) (2.2)
L2 \4lJ

For our three

CE : GM : ,[4+ = 1, for Dick (2.3)

y 4
CE : HM : f1/1\*16ayl-1: 1-2.' forHarrier (2.4)
t '' 1'7
!'to _1,
CE = KM : l:r\.:,/+l- : r+2, fsr Tom
2y4l 16'
(2.s)
12'

The lW formula of 1728Kramerdefinesfor Tom's CE a "root-squared

Mean", KM.

HM < GM < KM < AM : |f+>.t1-11 (2.6)

Dick is a pushyguy. Whatif he persuades Harrietto replaceher x* : 3 Uy his x* : !? If she

agreesto shootherselfin her own foot, the loss in her CE dollarsbelow her bestCE--dollars
is equivalentto her having agreedto throw awaya definablepercentageof her initial wealth.
"beingtrue
What'sleft, investedherproper way, will fall shortof what shecouldhavegot by
to her self" by a measurabledeadweightloss.

Persuasive $ harm to Tom if Tom givesup x* = 1.and goes

Dick could alsodo a measurable
alongwith Dick's x* : ]. A
A A ,lr
P^-*r /wL4'<L*k4- -a^fly''
r> A r a^Ä-
Lfu ,o r .c^i"J',Lfu'
.
AN--*
r ,^A n - ^ A ,
4u a$q'v"t
\ /
D"J^VÄjn {""/ A,sY",,*ttr 6* ) nDN2 " oslY'5rrr?

D- y-o-'w,"4^N {ÄfircüiF'
Mr. Elwyn Berlekamp
Page3
16 November2005

Cantheseone-periodharmserodeawayafter Tom andHarriet cometo shootthemselvesin their

respectiveiegs two times, threetimes, N : 1010times?No. No suchLimit Theoremis valid.
F o r N : u g e , N > > 1 , x * : f , a n d * - : * a n d x * : l e a c h p r o d u coen r e t i r e m e n t d a t e t h r e e
differentwide-spread Log Normallimit distributions.Tom'sLog Normalhasthe largestabsolute
arithmeticmeandollars.Harriet'shasthe leastabsolutearithmeticmeanof dollars.However,
at Harriet's requestwe calculatethe threeH.M.'s. Hers is the largest!
"dominating"retirementnest
Theorem: In no run, howeverlong, doesKelly's Rule effectuatea
c' bob 'o

In a vanity duel betweenany two neighbors,wherewhat's to be maximizediS A's probability

of being aheaAof B when they both retire at the sametime and startto invest at the sametime,
Dick rypeswill beatout both Harriet typesand Tom types.And for Methusala-ish neighbors,
Dick's probabüiryedgewill go to 1 ("almost"),as N * oo'

Elwyn, if I am wacky, you can set me straight. If I am right, that shouldn't invalidateany
importantpost-Shannon information-theoretictheorems.

Sincereiy,
r{} rq.
i ,// t i
{cr-ruUJ
PaulA. Samuelson

p.S. For two outcome stocks, solve for xn as the root of the first derivative equation

aJfUf:*+1)+1g/1-1*\I : 0. x* canbe foundfor thesethreeneighborsby solvingonly a

d x t 2 2 \ 4 l J
deducibleli.nearequation.That's only becauseall tfuee are speciesof the Club of Laplacians
//CW)^J/6$4 be a negative
with ConstarrtRelativeRisk Aversion, asdefinedby eachhaving WU
consmnt.
 - , /

MASSACHUSETTS
INSTITUTE OF TECHNOLOGY

D E P A R T M E N TO F E C O N O M I C S 50 MEMORIAL DRIVE
C A M B R I D G E ,M A S S A C H U S E T T S
02I42-1347

13 December2000

Professor William T. Ziemba

f lii;ei'sityof Brit;:hCalumbia
Facuityof Comnrerce
Vancouver, B.C. V6f 1ZZ
Canada

D e a rW i l l i a m ,

I cannot believeyour fecundity.Two huge books in the same post.

Thanks.And thanks.

You musttell me which threearticlescan be mostprofitableto me as

an investor.I collect"tips,"but from smartguysonly.

Best,

?^tb
PaulA. Samuelson

PAS/jmm

P.s. Croan.when I migratefor hibernationSaturday

to Florida,l,ll have
to portageyour two hugevolumes.

r
t
r
Fwd: markowitz chapter
Duo.:P*X
'tr)
Subject: Fwd:markowttzchapter Tloi s ÄnnesfespfrÄ Pnpen
From:WilliamZiembacwtzimi@mac.com)-fa'^
Date:Thu,08May 200809:38:48-0700 iil }',e*t;,Hur# l*
E,- n
To : JamesPoterba<poterba@MIT.EDU> I44€ I {.,s\ ,"*EA ,
li am Ziemb
C C : RacheI Ziemba <rachelziemb a@mac.c om>, William Ziemba ffi-:nnam\
<wrHrn
a <wtHmii@tna
@mac.com)
JI'P
you. PASis lUFp volume.IryX q
JIM wouldyoup1sprintfor P$S and;glfq$ if it interesls
iloieäL'Jä.*,
MW I ratherlike
t1
in here that Hauschand I did in 1986.t-Ii of independent
oÄ affufias.e ovET Fo medlum lensth on time/that alarge fraction of the time the
investormakeshuge gains. BUT its possibleto make 700 independentbets all wfih a 14oÄ
advantage with a decentchanceof winning eachone at oddsof 1-1 to 5-1 so probsof .19 to .51
AND LOSE 98% of onesoriginal wealth $ 1000 in our example
and fractionalkelly doesnot help much as the min:145 or 85.5% loss.
,lu$"#'{tä; ; {t# WW*a @"ffi'4.b4p4., .f t tf:.. V'
THE k;lli well 'in horseracingwherewe öancalculateexactslippage/see
app'roach-i%rks exatnplein A*ü"
paper/andthere are many many bets. HEDGE fund types like Jim Simons fu*"
rvho do a lot of similar bets have good successtoo. We are finding in a trend following fund
project that uncerataintiesin mean estimatesmakesit hard so far to beat i/n ex post/researchnot t*'vtL
done but 1/n is pretty fonlidable. ffi"
W*
BEST N ,
PS I worked a bit on your Hany, etc exampie for a talk I gave at U Chicagolast April.
{r{r"'gÄ'f*
I added? more investors IDA for IDA MAY FTILLER who was the first soc security recipient, b"1{
SHepaidin 24$ andlivedto be 100 andcollectedover24 000 Sireis risk aversionapproaching,Aj44
infinity being alphaWtothealphaforaiphagoing to -infinity ,.1., ,,1 :,1- ,,4t-
THENIon therisky side is VICTOR for VictorNiederhoffer who is linearutiiity with risk , r Ä^
aversion:0 THEN thereare5 with risk aversions infinity 2 1 ll2 antdzero. i{ÄWI
VICTOR goesbankruptfor sure as VN hasabout5 times/some of this about\N is discussed in a #iq1y'
chapterin the WILEY book I did with my daughter Zierha and.Ziemba 2007 Scenarios for risk
Md
managementandglobalinvestmentstrategies -
P \r
WE areplanninga kelly book reprintingthe classicarticles includingyour criticisms andthe great yY
resuits and commentariesSO I will draftsomething irere to respondto your 2 lettels We can M .
tlrendiscussthat then THx for your patienceon this üVrJ"*
,,/)4fi4"üÄ. .#r^*
Begin forwardedmessage:
*rn
t<"w*"
ge.ubc.ca>
F ro m : Wil 1iam Ziemb a <zi ernba(@interchan
Date: May 8, 2008 8:43:58AM PDT (CA)
U{ws-*
To : William Ziemba <wtzimi@lx ac.com> kEL'n""ffi
Cc:
"4+t4-ura
€ft#
I of 2 i 1:08PM
5/8/2008
 MASSACHUSETTS
INSTITUTE OF TECHNOLOGY

DEPARTMENT OF ECONOMICS

:Xffiil3äii,T:l'." "sErrso2142.|347
13 December2006

Professor William T. Ziemba

University of British Columbia
SauderSchool of Business
2053 Main Mall
Vancouver,BC Y6T lZ2
Canada

Dear William,

I received with pleasure a copy of your new 2006 edition of Stochastic Optimization Models in
Finance. Many thanks.

As I hastily turned its pages,I came to an over-hasty uneasethat maybe you are too kind to the
Kelly Criterion. (Probably I am the confusedone.)

This motivates me to send you a copy of some incomplete correspondence with Elwyn
Berlekamp, a Berkeley math professor who once was a Claud Shannon Ph.D. at MIT and who
has madebushelsof money as a trader and one-time activist in the Jim Simon RenaissanceGroup
(Medallion Fund, etc.). He is a good friend and a colleagueon the National Academy Finance
Committee.

I enclose a book review by him favorable to the Kelly criterion (which I think he related to a
valid Shannon theorem in modern communication theory).

"the money certainty-equivalent to a

I include my letter to him which used the concept of
stochasticportfolio" to measure how much of my fortune would be needlessly lost if I am a
Laplacian with non-log utility (such as 1@lt[ or -llwealth) and became seducedby Latanö-
Kelly-Breiman dogma. I deemed my arguments fatal to Kelly zealots. (However, the debate
remainedmoot. Elwyn never answeredmy letter.)

Maybe I can be luckier with you?

Enjoy,

P"'^L
PaulA. Samuelson

PAS/jmm
 I

MASSACHUSETTS
INSTITUTE OF TECHNOLOGY

DEPARTMENT OF ECONOMICS
5OMEMORIAL DRIVE
C A M B R I D G E , M A S S A C H U S E T T S0 2 1 4 2 . 1 3 4 7

7 May 2007

ProfessorWilliam Ziemba
SauderSchool of Business
University of British Columbia
2053 main Hall
Vancouver
British Columbia V6T lZ2
Canada

Dear William,

I believe I sentyou a letter spelling out in exactdetail how a very risk-averseLaplacian--callher

Harriet--would be throwing away a computable number of her initial wealth if, instead of
choosing a period-for-periodHarmonic Mean (called for by her degreeof risk aversion),she
instead usedyour Kelly criterion and maximized Kelly's Geometric Mean.

Equally fatal would it be for Tom, a Laplacian more risk tolerant than Kelly. Tom maximrzes
1728 Kramer Weäm-Üdlry. If investingfor one period or 1010periods,Tom were to use
Kelly's log Utility, I calculated for you exactly how much of his initial wealth he is flushing
down the toilet.

I still have been expecting your reply. Capitulation. Or cogent repudiation of my erroneous
deductions.Insteadyou only sentme a copy of Jim Thorpe's article in your anthology.I leafed
through it and threw it away. Subsequentlya Nobel Prize winner sent me a copy of it, saying,
"Seventyeleven times, Paul, you killed off that dragon.Yes. Thorpe made good money in Las

Vegas and in Princeton. But what test is that of the 1ogW Kelly criteria vs. the fiV tom or the
Harriet -W-1criteria?"

Get serious.

Amicably yours,

%^q-
Paul A. Samuelson

PAS/jmm
r

MASSACHUSETTS
INSTITUTE OF TECHNOLOGY

D E P A R T M E N TO F E C O N O M I C S
'O MEMORIALDRIVE
C A M B R I D G E ,M A S S A C H U S E T T S
02142.1347

L6 November2005

ProfessorElwyn Berlekamp
2039 ShatfuckAvenue
Berkeley,CA 94704

Dear Elwyn,

This first of two lettersthanksyou for the mailing relevantto the NAS's choiceamong
four optionsfor paying Renaissance management fees. What the Committeemust bring
to your deductivediagramsis its best guessasto how the newfund will do in the future
and how the S&P index is likely to do.

Your hunchwill probablybe nearerthe mark than that of the rest of us becauseof your
greater past experiencewith the Renaissancegroup. ftnO t suspectthat their past
experiencewith you may havehelpedinduceJim Simonto let us in at half the usual$20
million ante.So bravo!

My other letter, drafted earlier, arisesfrom my having been sent your valuablebook
review in the American Scientistof the recentPoundstonebook. I encloseif for your
possibleinterest,and I would benefit a lot from learningaboutany non-optimalitiesin
my analyses.

Cordially,

?Gtr^A.
PaulA. Samuelson
 M A SS A C H U S E T T SI N S T i T U T E O F T E C H N O L O G Y

D E P A R T M E N TO F E C O N O M I C S
5 OM E M O R I A L D R I V E
CÄMBRIDGEM , A S S A C H U S E T T0S2 14 2 . 1 3 4 7

16 November2005

ProfessorElwyn Berlekamp
2039 ShattuckAvenue.#408
Berkeley,CA 94704

Dear Elwyn,

I'm gratefulfor the reprint of your recent fineAmericanScientislbook review. .Exceptfor your
thoughtfulness, I'd haveiikely not learnedaboutit.
"rational"neighbors:Tom, Dick and
Hereis a fablethatmight be of someinterest.I havethree
Harriet. All are risk averterswho shunall "unfai.r"gambles.Being of unequaldegreeof risk
aversion,Harriet is the most cautious;Dick (Goldilocks)hasmore risk tolerancethan Harriet
but lessrisk tolerancethan sportiveTom. All faceone seriousproblem.They must investfor
a future period of retirementwhiie the feasiblechoiceis between(1) safecashwith only zero
yield, and (2) one singlestockthat in everyperiodwill, for each$1 investedin it now, bring
to the investorone period from now either $4 or $j at even odds.

"If
Testl. I testmy neighborin variousways. you mustput I00Voof your nesteggin only one
of thesetwo options,wltich will you pick?" Tom replies: 700%in the stock,x : 1; Harriet
replies:IA07oin safecash.1-x : l. Dick says:I'm inffirent betweenall in the volatilestock u ,-t _frfi
(n*"-fl* rGd
andatlin thesafecash. g*t-*.-) D;,l- äa* tc!$e.,-üüfu/h"- ff*q,)"Ä(""

Q. Given a horizon of N ) 1 periodsuntil the final dateof your retfuement,will you change
yow x proportions? To the surpriseof lay people,all threeanswerNo, no change.

Test II. Now you're giventhe optionof btendingcashandstock.Harriet replies:for everyN,

large or small,I'11put = ** in the stockand 7* = L-x * in safecash.No surprisewhen
X.
Tom reportsa larger x1 than Harriet: for Torn, x* : l.* and L-x- : 0*. Middling Dick, as
expected,hashis x* betweenthoseof the othertwo. For Dick, ** : ** : 1-x*.(MaybeDick's
last nameis Kelly or Breimanor Latan6.)

Why theseparticulardecisions?Interrogationsrevealthat all three are devoutLaplacians,who

ever strive toward maximal ExpectedConcaveUtility of wealth outcomes.

yii,r,ot.i"(i)I inrestI, u/ > o > u// ( 1 .1 )

Mr. Elwyn Berlekamp
Page2
16 November2005

-x).}u(}*.t -*)}, t restII.

vtax{}uC+x+1 (r.2)

Dick reportshis U(W) - log W (what economistscall 1738D. Bernoulliutility). What's the
samething, Dick is a GeometricMean maximizer

CautiousHarriet is a HarmonicMeanmaximizer,with U : -W-1*A. Her limited degreeof risk

tolerancefits pretty well lots of empirical Wall Street"equify premia" data.

To measurehow mucheachpersongainsor losesin (subjective)"CertaintyEquivalentDollars,"

I define their three different CE's by the following generalformulas:

u(cE): iu@).i"(*) (2.1)

cE : u -'[jut+1.;u(ä]= E(4,r) (2.2)

For our three

CE:GM: ^[4J = 1. for Dick (2.3)

Y 4
CE:HM: : 1-], torHarriet (2.4)
fi(i)-ir*i]-1
cE : KM : ltuq.:-,El'
2\ 4)
: L+2. forTom (2.s)
12' 16'

Th. /W formula of 1728Kramerdefinesfor Tom's CE a "root-squared

Mean", KM.

HM < cM < KM < AM : rr<o>.iä\, (2.6)

Dick is a pushy guy. What if he persuad.esHarriet to replaceher x* : ? by his x* : +? If she

agreesto shoothersettin her own foot, the loss in her CE dollars Uetiw n., b.rt CE'. dollars
is equivalentto her having agreedto throw away a definablepercentageof her initial weaith.
What's left, investedher proper way, will fall short of what shecould have got by "being true
to her self" by a measurable deadweightloss.

PersuasiveDick could also do a measurable$ harm to Tom if Tom gives up x* = 1 and goes
alongwith Dick's x. : *.
Mr. Elwyn Berlekamp
Page3
16 November2005

Cantheseone-periodharmserodeawayafter Tom andHarriet cometo shootthemselvesin their

respectivelegstwo times, threetimes,N = 1010times?No. No suchLimit Theoremis valid.
For N large, N > > 1, x* : and*- : j and x* : 1 eachproduceon retirementdate three
f,
differentwide-spreadLog Normallimit distributions.Tom's Log Normal hasthe largestabsolute
arithmeticmeandollars. Harriet's hasthe leastabsolutearithmeticmeanof dollars. However,
at Harriet's requestwe calculatethe threeH.M.'s. Hersis the iargest!

Theorem: In no run, howeverlong, doesKelly's Rule effectuatea "dominating"retirementnest

egg.

In a vanify duel betweenany two neighbors,wherewhat's to be maximizedis'A's probability

of being aheadof B when they both retire at the sametime and startto invest at the sametime,
Dick types will beat out both Harriet types and Tom types. And for Methusala-ishneighbors,
Dick's probabilifyedgewill go to 1 ("almost"),as N -- oo.

Elwyn, if I am wacky, you can set me straight. If I am right, that shouldn't invalidate aly
importantpost-Shannoni nformation-theoretictheorems.

Sincerely,

fre^-ü,
PaulA. Samuelson

P.S. For two outcome stocks, solve for x* as the root of the first derivative equation
g
11u1:**t)*1u(t-3r)I : 0. x* can be found for thesethree neighborsby solvingonly a
dxtz 2 \ + lJ
deduciblelinear equation.That's only becauseall three are speciesof the Club of Laplacians
with ConstantRelativeRisk Aversion,asdefinedby eachhaving WU /(W)^J/6W; be a negative
constant.