Theory
David Ellerman
World Bank
December 2000
Table of Contents
Introduction: Finding the arkets in the ath
!rbitrage in "ra#h Theor$
E%am#les of !rbitrage&Free Conditions
'irchhoff(s )oltage *a+
!ssemblies of "ears or Wheels
Cli,ue Formation in ocial "rou#s
.eat !rbitrage in Thermod$namics
!rbitrage in /robabilit$ Theor$
!rbitrage and 0#timi1ation Theor$
!rbitrage&Free Conditions on arket "ra#hs
!n Economic Inter#retation of Cofactors2 Determinants2 and Cramer(s 3ule
!rbitrage&Free arket atrices
First&0rder 4ecessar$ Conditions as !rbitrage&Free Conditions
3eferences
!##endi%: econd&0rder Conditions as !rbitrage&tabilit$ Conditions
Introduction: Finding the Markets in the Math
0ne of the fundamental insights of mainstream neoclassical economics is the connection bet+een
com#etitive market #rices and the *agrange multi#liers of o#timi1ation theor$ in mathematics5
6et this insight has not been +ell develo#ed5 In the standard theor$ of markets2 com#etitive
#rices result from the e,uilibrium of su##l$ and demand5 But in a constrained o#timi1ation
#roblem2 there seems to be no mathematical version of su##l$ and demand functions so that the
*agrange multi#liers could be seen as e,uilibrium #rices5 .o+ can one 7Find the markets in the
math7 so that *agrange multi#liers +ill emerge as e,uilibrium market #rices8
We argue that the solution to the 7Finding the markets in the math7 #roblem is to reconce#tuali1e
e,uilibrium as the absence of #rofitable arbitrage instead of the e,uating of su##l$ and demand5
With each #ro#osed solution to a classical constrained o#timi1ation #roblem2 there is an
associated market5 The ma%imand is one commodit$2 and each constraint #rovides another
commodit$ on this market5 "iven a marginal variation in one commodit$2 one can define the
marginal change is an$ other given commodit$ so the market has a set of e%change rates bet+een
the commodities5 The usual necessar$ conditions for the #ro#osed solution to solve the
ma%imi1ation #roblem are the same as the conditions for this mathematicall$ defined 7market7 to
be arbitrage&free5 The #rices that emerge from the arbitrage&free s$stem of e%change rates
9normali1ed +ith the ma%imand as numeraire: are #recisel$ the *agrange multi#liers5 We also
sho+ the cofactors of a matri% describing the marginal variations can be taken as the #rices
9before being normali1ed: so the *agrange multi#liers can al+a$s be #resented as ratios of
cofactors5
tarting +ith an$ s,uare m m matri% 9+ith rank m;<:2 a market can also be defined and the
cofactors given a #rice inter#retation so that an economic inter#retation can be constructed for
the inverse matri% and for Cramer(s 3ule5
The relevant mathematical result2 +hich dates back to !ugustin Cournot in <=>=2 is that:
there e%ists a s$stem of #rices for the commodities such that the given e%change
rates are the #rice ratios if and onl$ if the e%change rates are arbitrage&free 9in the
sense that the$ multi#l$ to one around an$ circle:5
This sim#le gra#h&theoretic theorem is kno+n in its additive version as 'irchhoff(s )oltage *a+
9')*::
there e%ists a s$stem of #otentials at the nodes of a circuit such that the voltages
on the +ires bet+een the nodes are the #otential differences if and onl$ if the
voltages sum to 1ero around an$ c$cle5
'irchhoff(s +ork +as #ublished in <=?@2 so it might be called 7the Cournot&'irchhoff la+57
There is also an additive version of the additive ')*5 If t+o commodities are s+a##ed2 one unit
for one unit2 then usuall$ some additional 7boot7 must be #aid for the higher valued commodit$5
For each #air of goods i and A2 su##ose +e are given an amount boot9i2A: that is the additional cash
boot that needs to be #aid along +ith one unit of good i in order to receive one unit of good A5
Then ')* takes the form:
"iven a s$stem of boots for commodit$ s+a#s2 there e%ists a set of unit #rices for
the goods such that the boot necessar$ for an e%change of units is the #rice
difference if and onl$ if the s$stem of boots is arbitrage&free in the sense of
summing to 1ero around an$ circle5
We sho+ that this Cournot&'irchhoff la+ has man$ a##lications outside of electrical circuit
theor$ and economics5 For instance2 the second la+ of thermod$namics can be formulated as the
im#ossibilit$ of a certain form of 7heat arbitrage7 bet+een tem#erature reservoirs2 and the
7#rices7 that emerge in this case are the 'elvin absolute tem#eratures of the reservoirs5 6et
another a##lication of the arbitrage frame+ork is in #robabilit$ theor$5 /rofitable arbitrage in the
market for contingent commodities is called 7making book57 ! #erson(s subAective #robabilit$
Audgments satisf$ the la+s of #robabilit$ if the$ are 7coherent7 in the sense of not allo+ing book
to be made against the #erson5 Thus arbitrage on the market for contingent commodities enforces
the la+s of #robabilit$5
!rbitrage&related conce#ts have been a##lied successfull$ in financial economics5 erton .5
iller and Franco odigliani used im#ressive arbitrage arguments in #roving their famous
irrelevance theorem B<CD=E5 te#hen !5 3oss B3oss <C@Fa2 <C@FbE and his colleagues have
develo#ed !rbitrage /ricing Theor$ so that it is no+ recogni1ed as a fundamental #rinci#le in
finance theor$ B)arian <C=@E5 0ur #ur#ose here is not to use arbitrage conce#ts to stud$ financial
markets2 but to find the mathematicall$ defined 7markets7 and the related arbitrage&conce#ts in
the mathematics of all classical constrained o#timi1ation #roblems5
2
Arbitrage in Graph Theory
! directed graph " G9"
0
2"
<
2 t2h: is given b$ a set "
0
of nodes 9numbered 02<25552m:2 a set "
<
of
arcs 9numbered <2225552b:2 and head and tail functions h2t:"
<
"
0
2 +hich indicate that arc A is
directed from its tail2 the t9A: node2 to its head2 the h9A: node5
t9A: h9A:
!rc A
Figure 1. !rc A from Tail t9A: to .ead h9A:
It is assumed that there are no loo#s at a node2 i5e52 h9A: t9A: for all arcs A5 ! path from node i to
node i( is given b$ a se,uence of arcs connected at their heads or tails that reach from node i to
node i(5 ! gra#h is connected if there is a #ath bet+een an$ t+o nodes5 It is assumed that the
gra#h " is connected5 ! closed circular #ath +here no arc occurs more than once is a cycle Bfor
more gra#h theor$2 see an$ te%t such as Berge and "houila&.ouri <CFDE5
*et T be an$ grou# 9not necessaril$ commutative: +ritten multi#licativel$ 9i5e52 a set +ith a binar$
#roduct o#eration defined on it2 +ith an identit$ element < and +ith ever$ element having a
multi#licative inverse or reci#rocal:5 For most of our #ur#oses2 T can be taken as 3H2 the
multi#licative grou# of non1ero reals5 In the motivating economic inter#retation2 a different
commodit$ is associated +ith each node2 and the arcs re#resent channels of e%change or
transformation bet+een the commodities at the nodes5 ! function r:"
<
T is a rate system giving
e%change or transformation rates5 "iven an arc A2 one unit of the t9A: commodit$ can be
transformed into r9A: G r
A
units of the h9A: commodit$5
t9A: h9A:
!rc A
3ate r
A
Figure 2. Transformation 3ate r
A
on !rc A
! gra#h 9"2r: +ith a rate s$stem r re#resents a market2 so it +ill be called a market graph. These
grou#&labeled gra#hs are also called 7voltage gra#hs7 B"ross <C@?E or 7grou# gra#hs7 B.arar$ et
al5 <C=2E5
!ll transformations are reversible5 If arc A is traversed against the arro+2 the transformation rate
is the reci#rocal <Ir
A
5 "iven a #ath c from node i to i(2 the composite rate rBcE is the #roduct of the
rates along the #ath using the reci#rocal rate for an$ arc traversed against the direction of the
arro+5 ! function /:"
0
T labeling the nodes is a price system 9or absolute #rice s$stem:5 !
rate s$stem J9/::"
<
T can be derived from a #rice s$stem b$ taking the #rice ratios
J9/:9A: G /9h9A::
;<
/9t9A::5
Equation 1. Derived 3ate on arc A G /rice at Tail Divided b$ /rice at .ead
Derived rate s$stems have certain s#ecial #ro#erties:
<5 for an$ #ath c from i to i(2 J9/:BcE G /9i(:
;<
/9i:2
>
25 for an$ t+o #aths c and c( from i to i(2 J9/:BcE G J9/:Bc(E2 and
>5 for an$ c$cle c2 J9/:BcE G <5
"iven a market gra#h 9"2r:2 the rate s$stem r is said to be pathindependent if for an$ t+o #aths c
and c( bet+een the same nodes2 rBcE G rBc(E5 The rate s$stem is said to be arbitragefree if for an$
c$cle c2 rBcE G < B7arbitrage&free7 G 7balanced7 in much of the gra#h&theoretic literature follo+ing
.arar$ <CD>E5
r G <I2
r G >
r G >I2
r G ?I>
r G ?
r G <I<2
# G < # G 2
# G F
# G ?
# G >
# G <2
rBcE G9<I2:9<I>:9>I2:9?I>:9<I?:9<2: G <
Figure 3. !n !rbitrage&Free arket "ra#h
In an ideali1ed international currenc$ e%change market +ith no transaction costs2 if the #roduct of
the e%change rates around a circle is greater than one2 #rofitable arbitrage is #ossible5 If the
#roduct is less than one2 then e%change around the circle in the o##osite direction +ould be
#rofitable arbitrage5 .ence the market is arbitrage&free if the #roduct of e%change rates around
the circle is one5
! rate s$stem derived from a #rice s$stem has both the #ro#erties of being #ath&inde#endent and
arbitrage&free2 and2 in fact2 the three #ro#erties are e,uivalent5 That e,uivalence theorem is the
finite multi#licative version of the calculus theorem about the e,uivalence of the conditions:
<5 a vector field is the gradient of a #otential function2
25 a line integral of the vector field bet+een t+o #oints is #ath&inde#endent2 and
>5 a line integral of the vector field around an$ closed #ath is 1ero5
Cournot!irchhoff Arbitrage Theorem: *et 9"2r: be a market gra#h +ith r:"
<
T taking
values in an$ grou# T5 The follo+ing conditions are e,uivalent:
<5 there e%ists a #rice s$stem / such that J9/: G r2
25 the rate s$stem r is #ath&inde#endent2 and
>5 the rate s$stem r is arbitrage&free5
For a #roof of this straightfor+ard noncommutative generali1ation of 'irchhoff(s )oltage *a+
9<=?@: and Cournot(s earlier 9<=>=: arbitrage&free condition2 see Ellerman B<C=?2 <CC0E5
?
"#amp$es of ArbitrageFree Conditions
!irchhoff%s &o$tage 'aw
The original 7arbitrage&free7 condition in electrical circuit theor$ is 'irchhoff(s voltage la+
9')*:5 It is the additive version of the multi#licative arbitrage #rinci#le5 In economics2 the
commodit$ +ith the #rice of < is the numeraire5 In circuit theor$2 the node +ith a #otential of 0 is
the 7ground7 or 7datum7 node5 ! real&valued function on the nodes of a gra#h is a 7#otential57
The additive version of the ,uotient o#erator J9: is the difference o#erator2 +hich assigns to each
arro+ the difference bet+een the #otentials at the tail and head of the arro+5 If an assignment of
reals to the arro+s of the gra#h comes from a #otential on the nodes b$ taking these differences2
then the assignment to the arro+s is called a 7#otential difference7 or 7tension57 3eals assigned to
the arro+s can be added u# around an$ c$cle 9taking care to take the negative of the number if
the arro+ is traversed back+ards:5 The !rbitrage Theorem then $ields:
')*: !n assignment to the arro+s is a #otential difference if and onl$ if it adds
to 1ero around an$ c$cle5
Assemb$ies of Gears or (hee$s
! train of gears 9or +heels: that +ent around in a circle +ould be #erfectl$ useless2 but it #rovides
an amusing e%am#le of an arbitrage&free condition5 "ear ratios multi#l$ along a gear train so this
e%am#le uses the arbitrage theorem in its multi#licative form5 !ngular velocities on the shafts
#la$ the role of the commodit$ #rices5 If angular velocities can be assigned to the shafts so that
their ,uotients are the gear ratios2 then the +hole gear assembl$ can move5 0ther+ise it +ould be
rigid5 Thus a gear assembl$ has a motion if and onl$ if the #roduct of gear ratios around an$
circular gear train is one5
B$ #lacing t+o or more gears on the same shaft2 a circular gear train need not have all the gears in
the same #lane5 But if all the gears are in the same #lace 9e5g52 if the$ are all l$ing on a table:2
then the #roduct of gear ratios around an$ circle +ill al+a$s be #lus one 9even number of gears in
the circle: or minus one 9odd number of gears in the circle:5 Thus a circular gear train +ith all the
gears in the same #lane can move if and onl$ if it has an even number of gears5 "ra#hic artists
sometimes dra+ a sim#le #icture of three gears meshing in a circle2 and some organi1ations have
even used such an image as their logo5 But such a gear train is a #erfect e%am#le of gridlock
since it cannot move5
)
Figure 4. ! 3igid Circular Wheel !ssembl$
D
C$i*ue Formation in +ocia$ Groups
The arbitrage condition a##lied to 7likes7 and 7dislikes7 in social grou#s might give some insight
into the ethnic mentalit$ +here likes and dislikes are based largel$ on being inside or outside of
the cli,ue2 clan2 or tribe5 Each node in the gra#h is a #erson and each arro+ has K< or ;<
according to +hether the #erson at the tail of the arro+ likes or dislikes the #erson at the head of
the arro+5 Then a gra#h is said to be 7balanced7 if it is arbitrage&free in the sense of the likes and
dislikes multi#l$ing to K< around an$ circle Be5g52 .arar$ <CD>2 .arar$2 4orman and Cart+right
<CFDE5 The classic 7mother&in&la+ triangle7 is an e%am#le of an unbalanced gra#h5
other&in&*a+
.usband Wife
K<
K<
&<
Figure . !n 7Lnbalanced7 ocial "rou#
! 7#rice s$stem7 marks each node or #erson +ith K< or ;<2 and a given #attern of likes and
dislikes is derived from such a marking if each #erson likes others +ith the same marking and
dislikes those +ith a different marking5 Then the arbitrage theorem gives the follo+ing result5
! social grou# +ith a given #attern of likes and dislikes can be #artitioned into t+o
clans such that all likes are intraclan and all dislikes are bet+een clans if and onl$ if
the #attern of likes and dislikes is balanced 9arbitrage&free:5
Thus there is no +a$ to grou# the three #eo#le in the mother&in&la+ triangle into t+o families to
account for the likes and dislikes5 The husband and mother&in&la+ 9+ife(s mother: have to be in
different families to account for their dislike2 but then the +ife has an identit$ crisis5 When
arbitrage is #ossible then2 in effect2 a commodit$ has t+o #rices 9so one can bu$ lo+ and sell
high:5 In the #revious e%am#le2 a +heel had to rotate in t+o directions at once in order for the
+heel assembl$ to move5 In this e%am#le2 the #attern of likes and dislikes in the mother&in&la+
triangle #uts the +ife in the #osition of having t+o conflicting famil$ identities5
,eat Arbitrage in Thermodynamics
The Carnot engine a##roach to the second la+ of thermod$namics 9sim#lified for a finite number
of tem#eratures: gives an a##lication of the arbitrage theorem in #h$sics5 Each node is a heat
reservoir +ith a different tem#erature 9including for calibration #ur#oses the free1ing and boiling
#oints of +ater:5 Each arro+ is a Carnot engine that can reversibl$ +ithdra+ the heat dJ
c
from
the lo+&tem#erature reservoir b$ #erforming the +ork dW2 and dum# the heat dJ
h
into the hotter
reservoir +here dJ
h
G dW K dJ
c
b$ the first la+ of thermod$namics 9conservation of energ$:5
The ratio r G dJ
c
IdJ
h
is called the efficiency debit and is the #ositive real number assigned to the
arro+5 When Carnot engines are hooked in series2 the com#osite efficienc$ debit is the #roduct of
the efficienc$ debits of the individual engines5
F
.ot
Cold
dJ
h
dJ
c
dW dJ
h
dJ
c
r G I
Figure !. ! Carnot Engine
0ne formulation of the second la+ of thermod$namics is that bet+een an$ t+o tem#eratures2
there is #ath inde#endence in the sense that the various connecting #aths must have the same
efficienc$ debit Be5g52 orse <CF?2 D0E5 0ther+ise one could #erform a t$#e of 7heat arbitrage7
9move heat from the cold to hot reservoir +ith no net e%#enditure of energ$: and have a
7#er#etual motion machine of the second kind7 Be5g52 Castellan <CF?2 Cha#ter =E5 B$ the
arbitrage theorem2 the second la+ im#lies that there e%ists a thermod$namic 7#rice7 T at each
node or reservoir such that the efficienc$ debit of each Carnot engine is the 7#rice ratio7
T
cold
IT
hot
5 If +e normali1e the free1ing #oint of +ater to 0 and the boiling #oint to <002 then the
7#rices7 are the 'elvin absolute tem#eratures of the reservoirs5
Arbitrage in robabi$ity Theory
7aking book7 means making a series of bets so that one has #ositive net earnings no matter
+hat ha##ens5 That is e,uivalent to #erforming #rofitable arbitrage on the market for contingent
commodities5 ! contingent commodit$ is a commodit$ conditioned on the occurrence of an
event2 e5g52 M<000 if $our number comes u# in a lotter$5 ! #erson subAectivel$ assigns a
#robabilit$ #9E: to an event E if the #erson is Aust +illing to #a$ #9E: in order to receive the
stake  if the event E occurs5 Thus #9E: is the #rice the #erson is +illing to #a$ for the
contingent commodit$ 7M< if E57 u##ose that a bettor #laces t+o bets +ith a bookie: the bettor
#a$s M< to get M2 if it is raining at noon2 and #a$s M<50D to get M2 if it is not raining at noon5 B$
taking both bets2 the bookie 7makes book57 4o matter +hat ha##ens2 the bookie gives u# M2 and
receives M250D 9G <500K<50D: for a net #rofit of M50D5 The bettor(s #robabilit$ assignments are
said to be coherent if book cannot be made against the bettor 9unlike the e%am#le:5 3amse$
B<CF0 9orig5 <C2F:E and de Finetti B<CF? 9orig5 <C>@:E sho+ed that the la+s of #robabilit$ theor$2
such as #9E: K #9not&E: G <2 could be derived from the re,uirement of coherence5 !rbitrage on
the market for contingent commodities enforces the la+s of #robabilit$5 Even if each #erson has
coherent #robabilit$ Audgments2 bookies can still make their living off the combined incoherence
of different #eo#le(s #robabilit$ Audgments5
Arbitrage and Optimization Theory
! sim#le e%am#le of an o#timi1ation #roblem +ill no+ be used to illustrate our main to#ic2 the
inter#retation of the necessar$ conditions for o#timi1ation as an arbitrage&free condition5
u##ose that the #roblem is to find the #ro#ortions for a rectangular fenced field of ma%imum
area for a given cost +hen one length of the field re,uires a form of fencing costing four times the
fencing used on the other three sides5
@
M? #er foot
M< #er foot
on other three sides
Width
G W ft5
*ength
G * ft5
Figure ". a%imi1e 3ectangular !rea +ith "iven Cost
There are t+o commodities on the market2 cost dollars and s,uare feet of area5 There are t+o
+a$s to transform an e%tra dollar into area: s#end the dollar to increase the +idth of the field or
to increase the length of the field5 If the dollar is s#ent on the +idth2 then it bu$s an e%tra half
foot on the +idth 9the e%tra foot needs to be s#lit bet+een the t+o +idths to kee# the rectangular
sha#e: so the area goes from W* to 9W K <I2:*5 .ence the e%tra area is *I25 If the dollar is
s#ent on the length2 then onl$ one&fifth of a foot can be added to the length 9M5=0 for one&fifth
foot on the e%#ensive side and M520 for one&fifth foot on the chea# side:5 Thus the area is
increased from W* to W9* K <ID: and the e%tra area is WID5 .ence there are t+o e%change rates
from the cost dollars to the s,uare feet of area are *I2 and WID5 This 7market7 can be #ictured in
an 7arbitrage diagram57
Cost Dollars ,uare Feet of !rea
*I2
WID
Figure #. !rbitrage Diagram for a%imum !rea /roblem
This market is arbitrage&free if and onl$ if the t+o e%change rates bet+een the commodities are
e,ual: *I2 G WID5 .ence the ma%imum area field is obtained +hen the length is t+o&fifths or ?0
#ercent of the +idth5 When formulated as a constrained ma%imi1ation #roblem 9ma%imi1e area
subAect to a fi%ed cost:2 that common rate *I2 G WID is the *agrange multi#lier for the #roblem5
ArbitrageFree Conditions on Market Graphs
The value grou# T +ill no+ be s#eciali1ed to 3H2 the multi#licative grou# of non1ero real
numbers5 But #rice s$stems / +ill no+ be e$tended b$ allo+ing 1ero values in the reals 32 i5e52
/:"
0
35
!n e%tended #rice s$stem / and a rate s$stem r are associated if for an$ arc A2
/9h9A::r
A
; /9t9A:: G 05
=
If the #rice s$stem has all non1ero values2 this is the same as the rate s$stem being derived from
the #rice s$stem5
The 1ero&#rice s$stem 9all 1ero #rices: is triviall$ associated +ith an$ rate s$stem5 If a rate s$stem
is not arbitrage&free2 then the 1ero&#rice s$stem is the onl$ associated #rice s$stem5 With that
fi%ed&rate s$stem2 #rofitable arbitrage means 7getting something for nothing27 so all commodities
become free goods and have 1ero #rices5
It is useful to reformulate some of the gra#h&theoretic notions using incidence matrices5 "iven
9"2r:2 the nodearc incidence matri$  G B
iA
E is the 9mK<: b matri% +here:

iA
+ r
A
if
!rc A
< if
!rc A
0 0ther+ise
'
4ode i
4ode i
Equation 2. 4ode&!rc Incidence atri%
The A
th
column of  has a minus one 9;<: and a rate r
A2
+hich are the results of transforming one
unit of the t9A: good into r
A
units of the h9A: good5 !n$ linear combination of the columns +ould
re#resent a #ossible market&e%change vector using the rate s$stem r5 The negative com#onents
re#resent the goods given u# in e%change for the goods re#resented b$ the #ositive com#onents5
Thus the vector s#ace of all linear combinations of columns of 2 the column space Col9:2 +ill be
called the e$change space of the market gra#h 9"2r:5
*et 
0
2 called the reduced incidence matri$2 be the m b matri% obtained from  b$ deleting the
to# ro+2 the ro+ corres#onding to node 05 If " is a connected gra#h 9a #ath bet+een an$ t+o
nodes:2 then the reduced incidence matri% 
0
has linearl$ inde#endent ro+s2 i5e52 
0
has full ro+
rank5 *et /H G 9/
<
25552/
m
: be a ro+ vector such that /H
0
G 05 ome node i +as connected to
the 7deleted7 node 0 b$ some arc A5 In order for /H to 1ero the A
th
column of 
0
2 /
i
must be 1ero5
If arc A is from node i to i( both in the node set Ni25552mO2 then /H
0
G 0 im#lies /
i(
r
A
; /
i
G 0 so /
i(
and /
i
are both 1ero or both non1ero5 Thus each node connected to node i must have a 1ero
#rice5 ince " is connected2 all #rices must be 1ero2 i5e52 /H G 02 so the ro+s of 
0
are linearl$
inde#endent5
!dding back the to# ro+2 the ro+ rank of  is either m or mK<2 so the column rank2 i5e52 the
dimension of Col9:2 is also either m or mK<5 ! subs#ace of 3
mK<
of dimension m 9one less than
the dimension of the full s#ace: is a hyperplane through the origin5 Thus the e%change s#ace is
either a h$#er#lane in 3
mK<
or is the full s#ace5
The left nullspace *eft4ull9: of an$ matri%  is the s#ace of vectors / such that / G 05 If  is
the incidence matri% of a market gra#h 9"2r: and / G 9/
0
2/
<
25552/
m
: is in *eft4ull9:2 i5e52 / G 02
then for all arcs A
/
h9A:
r
A
; /
t9A:
G 0
so / is a #rice s$stem associated +ith the rate s$stem r5 .ence *eft4ull9: is called the price
space associated +ith the e%change s#ace Col9: and the elements / are called price vectors5 The
C
e%change s#ace Col9: and the #rice s#ace *eft4ull9: are orthogonal complements of one
another2 i5e52
a5 P is an e%change if and onl$ if for an$ #rice vector /2 /P G 02 and
b5 / is a #rice vector if and onl$ if for an$ e%change P2 /P G 05
ince the$ are orthogonal com#lements2 dimBCol9:E K dimB*eft4ull9:E G mK<5 ince the
e%change s#ace is of dimension m or mK< 9" is assumed connected:2 the dimension of the #rice
s#ace is either one or 1ero5 ! #rice vector +ith an$ non1ero com#onents must have all non1ero
com#onents5 !n$ t+o non1ero #rice vectors must be scalar multi#les on one another5 The t+o
cases of a one or 1ero dimensional #rice s#ace corres#ond to the cases of 9"2r: being arbitrage&
free or allo+ing #rofitable arbitrage5 If #rofitable arbitrage is #ossible2 then the fi%ed non1ero
e%change rates r +ould allo+ one to generate an$ ,uantities of the goods so all commodities are
free goods2 i5e52 / G 0 is the onl$ #rice vector5 These results and some eas$ conse,uences are
collected together in the follo+ing theorem5
Arbitrage Theorem for Market Graphs: *et 9"2r: be a market gra#h +here " is connected
and r:"
<
3H5 The follo+ing conditions are e,uivalent:
<5 there e%ists a #rice s$stem /:"
0
3H such that J9/: G r2
25 the rate s$stem r is #ath&inde#endent2
>5 the rate s$stem r is arbitrage&free2
?5 the #rice s#ace *eft4ull9: is one&dimensional2
D5 the e%change s#ace Col9: is a h$#er#lane 9+ith a non1ero #rice vector as a normal vector:2
F5 the to# ro+ of 2 s
0
2 can be e%#ressed as a linear combination of the bottom m ro+s 
0
of 2
i5e52 there e%ist Q G 9Q
<
25552Q
m
: such that s
0
K Q
0
G 02 and
@5 if an e%change vector b G % has b
<
G 555 G b
m
G 02 then b
0
G 05
The incidence&matri% treatment of market gra#hs suggests a generali1ation of the economic
inter#retation to a more general matri% conte%t5 The ro+s re#resent commodities5 The columns
s#ecif$ e%change or #roduction #ossibilities5 4egative entries re#resent goods given u# in
e%change or in#uts to #roduction2 +hile #ositive com#onents stand for goods received or the
out#uts5 !n$ scalar multi#le2 #ositive or negative2 of a column also re#resents a #ossible
e%change or transformation so the column s#ace is the s#ace of #ossible e%changes or
transformations5 The orthogonall$ com#lementar$ left nulls#ace is the set of #rice vectors such
that all the e%changes can be obtained as trades at those #rices Bfor more linear algebra2 see an$
te%t such as trang <C=0E5
An "conomic Interpretation of Cofactors. /eterminants. and Cramer%s 0u$e
*et ! be a s,uare 9mK<: 9mK<: of reals2 and let !9k: be the 9mK<: m matri% obtained b$
deleting column k for k G 02<25552m5 The column s#ace Col9!9k:: is the s#ace of e%changes
s#anned b$ the remaining m columns5 *et
/9k: G 9/
0
9k:2/
<
9k:25552/
m
9k::
<0
be the cofactors of the deleted column k5 B$ the #ro#ert$ of 7e%#ansion b$ alien cofactors27
/9k:!9k: G 0 so /9k: is a 7#rice vector7 in *eft4ull9!9k::5 The cofactors in /9k: +ill be called
the kprices. The cofactors of an$ column of ! are #rices so that the e%changes defined b$ the
remaining columns can be obtained at those market #rices5
4o+ introduce the e%change 9or #roductive: #ossibilities given b$ the deleted column k into the
market5 Its value at the reigning #rices /9k: is the determinant R!R obtained b$ the cofactor
e%#ansion of column k5 If R!R 0 then an$ vector b can be obtained as an e%change vector
!% G b5 !s in a market that allo+s #rofitable arbitrage at fi%ed e%change rates2 an$ e%change is
allo+ed and the onl$ #rice vector is the 1ero vector5
It is therefore desirable to alter tem#oraril$ the inter#retation of the columns of !5 /reviousl$ the
columns re#resented e%change or #roduction #ossibilities +ith all commodities involved as in#uts
or out#uts listed as com#onents5 We no+ inter#ret each column as re#resenting the reversible
in#ut&out#ut vector of a machine o#erating at unit level5 But the machine(s services are not
re#resented in the in#ut&out#ut vector2 so the value of the vector can no+ be inter#reted as the
com#etitive rent im#uted to a unit of the machine services5
The vector of cofactor k&#rices /9k: G 9/
0
9k:2/
<
9k:25552/
m
9k:: can no+ be inter#reted as a set of
commodit$ #rices that im#ute 1ero rents to all the other m machines 9e%cluding the k
th
machine:5
The determinant R!R is the com#etitive rent 9or subsid$2 if negative: im#uted to the unit services
of machine k at those k&#rices5 Dividing b$ the determinant&as&rent2 the normali%ed kprices are
the k&#rices e%#ressed in terms of the units of machine k services as numeraire5
( )
( )
/ k
/ k
!
H
Equation 3. /rices to "ive Lnit 3ent to achine k2 Sero 3ent to 0ther achines
!t the normali1ed k&#rices /H9k:2 all machines have 1ero im#uted rentTsave machine k2 +hich
has an im#uted rent of unit$5 This $ields an economic inter#retation of the inverse matri$ !
;<
as
the normali%ed price matri$ obtained as the column of ro+ vectors /H9k: for k G 02<25552m5
( )
( )
( )
<
!
m H /
< H /
0 H /
H /
1
1
1
1
]
1
1
1
1
1
1
1
1
1
]
1
.ence the 9mK<: % 9mK<: matri% obtained b$ adding the 3. column vector as the mK<
st
column
is singular5 Thus its determinant obtained b$ the cofactor e%#ansion of the mK<
st
column is 1ero2
i5e52 /
0
Hr
i
; /
i
H G 0 so r
i
G /
i
HI/
0
H5
The ne%t theorem states a number of conditions e,uivalent to the market matri% ! being
arbitrage&free5 !n arbitrage&free market has uni,ue relative #rices so the C9n2m: local cofactor
#rices must mesh or fit together in the sense of being scalar multi#les of the non1ero #rice vector
/H +hich +as normali1ed to 9<2Q
<
25552Q
m
:5 The s#ace s#anned b$ the C9n2m: cofactor #rice
vectors is the one&dimensional s#ace *eft4ull9!:5 In the a##lication to classical o#timi1ation2 the
Q
i
(s are the *agrange multi#liers of m constraints2 +hich are thus inter#reted as the uni,ue #rices
of m resources in terms of the ma%imand as numeraire5
<>
!rbitrage Theorem for arket atrices: *et ! be an$ 9mK<: n market matri% +here +e
assume the ro+s < through m are linearl$ inde#endent5 *et a
0
be the to# ro+2 and let !
0
be the
bottom m ro+s of !5 The follo+ing conditions are e,uivalent:
<5 ! is arbitrage&free2
25 the #rice s#ace *eft4ull9!: is one&dimensional2
>5 the e%change s#ace Col9!: is a h$#er#lane 9+ith a cofactor #rice vector as a normal vector:2
?5 there e%ists Q G 9Q
<
25552Q
m
: such that a
0
K Q!
0
G 02 and
D5 if an e%change vector b G !% has b
<
G 555 G b
m
G 02 then b
0
G 05 Bee Ellerman <CC0 for the
#roof5E
FirstOrder 1ecessary Conditions as ArbitrageFree Conditions
The intuitive arbitrage reasoning as +ell as the formal results for arbitrage&free market matrices
can be a##lied to $ield the first&order necessar$ conditions for regular constrained o#timi1ation
#roblems +ith e,ualit$ constraints5
Consider the one&constraint #roblem:
a%imi1e $ G f9%
<
25552%
n
:
ubAect to: g9%
<
25552%
n
: G b
+here all functions are continuousl$ t+ice differentiable5 There are t+o commodities2 the
resource b and the ma%imand $5 There are n 7instruments7 +ith the levels of o#eration %
<
25552%
n
5
!t the levels %
<
25552%
n
2 the amount of the resource used&u# is g9%
<
25552%
n
:2 and f9%
<
25552%
n
: is the
amount of the ma%imand #roduced5
*et %
o
G 9%
<
o
25552%
n
o
: be levels of the instruments that use u# all of the available resource2 i5e52
g9%
<
o
25552%
n
o
: G b5 oreover2 +e assume that %
o
is 7regular7 in the sense that not all the #artials
g9%
o
:I%
i
G g
i
are 1ero5 We consider an intuitive 7marginal market7 defined b$ the #ossible
marginal transformations of b into $5 In an international currenc$ market 9+ithout transaction
costs:2 there might be n banks or e%change houses that to #revent arbitrage +ould have to offer
the same rate of e%change bet+een an$ t+o currencies5 In our market2 the n instrument variables
offer n +a$s to transform the resource b into the ma%imand $5 ! marginal variation %
i
uses&u#
g
i
%
i
units of b and #roduces f
i
%
i
units of $ so the rate of transformation is
b $
f %
g %
f
g
i i
i i
i
i
n 2 <
n 2 <
g g g
f f f
!
+here ;g
i
is used instead of Kg
i
since g9%
<
25552%
n
: re#resents the amount of the resource used u#5
Consider an$ column of this market matri% cou#led +ith the dumm$ column to form a s,uare
matri%:
8
8
5
f
g
i
i
1
]
1
The cofactors of the dumm$ column are the local #rices /
$
G ;g
i
and /
b
G ;f
i
2 so 9assuming
g
i
0: the cofactor #rice ratio is the transformation rate defined b$ the marginal variations in the
instrument %
i
5
b $
/ / f g
b $ i i
1
]
1
1
]
1
< 2
< 2
2 <
2 D
and the cofactor #rice ratios are given b$ the cofactors of the dumm$ columns in the matrices:
8
8
8
8
5
%
or
%
< 2
D 2
1
]
1
1
]
1
The mathematicall$ defined 7market7 is arbitrage&free if all the cofactor #rice ratios are the same:
/
/
% %
b
$
< 2
D 2
+hich gives the #revious necessar$ conditions that the length %
2
must be t+o&fifths or ?0 #ercent
of the +idth %
<
5
Consider a #roblem +ith m G 2 constraints:
a%imi1e $ G f9%
<
25552%
n
:
ubAect to: g
<
9%
<
25552%
n
: G b
<
g
2
9%
<
25552%
n
: G b
2
+here n V m G 25 *et " be the matri% of #artials of the constraints evaluated at %
o
:
5
g g g
g g g
"
2
n
2
2
2
<
<
n
<
2
<
<
1
]
1
1
]
1
1
< <
2 2
is nonsingular5 "iven the matri% 9+ith the unkno+n dumm$ column:
8
8
8
f f
g g
g g
i A
i A
i A
1
]
1
1
1
1
< <
2 2
the cofactors of the dumm$ column $ield the #rices:
/ g g g g
/ f g f g
/ f g f g
$ i A i A
b i A A i
b A i i A
< 2 2 <
2 2
< <
<
2
Equation 12. Cofactor /rices
+here /
$
0 b$ the choice of i and A5 B$ the cofactor #rice theorem2 the cofactor #rice ratios
$ield the transformation rates from the resources into the ma%imand5 For instance2 if %
i
and %
A
are
varied to hold b
2
constant2 the relative cofactor #rice of b
<
in terms of $2 /
b
<
I/
$
G Q
<
2 gives the
rate of transformation of b
<
into $ defined b$ the variation in %
i
and %
A
5
For the intuitive market to be arbitrage&free2 all the local cofactor #rices 9/
$
(2/
b
<
(2/
b
2
(: defined b$
an$ set of m G 2 instruments must be scalar multi#les of the non1ero #rice vector 9/
$
2/
b
<
2/
b
2
:5 In
formal terms2 the market matri% defined b$ the #roblem is
!
f
"
1
]
1
5
The first&order necessar$ conditions for the candidate #oint to be a constrained ma%imum are then
e%#ressed b$ the market matri% ! being arbitrage&free and b$ the other e,uivalent conditions
given in the !rbitrage Theorem for arket atrices5
!ll these results for m G 2 e%tend to the general #roblem +ith m constraints and n variables
9n V m::
<@
a%imi1e $ G f9%
<
25552%
n
:
ubAect to: g
<
9%
<
25552%
n
: G b
<
W
g
m
9%
<
25552%
n
: G b
m
5
The candidate #oint %
o
satisfies the constraints and is regular in the sense that the m n matri% "
G B g
A
i
E is of full ro+ rank5 Thus there are m columns forming a nonsingular submatri% "H5 If fH
is the vector of the corres#onding m #artials of f2 then consider the 9mK<: 9mK<: matri%:
8 H
8 H
5
f
"
1
]
1
The cofactors of the dumm$ column form the local cofactor #rices /
$
2/
b
<
25552/
b
m
determined b$
the m chosen instruments5 The intuitive market is arbitrage&free if all the C9m2n: vectors of local
cofactor #rices are scalar multi#les of this non1ero vector5 In formal terms2 the first&order
necessar$ condition for the candidate #oint %
o
to be a constrained ma%imum is e,uivalent to the
condition that the market matri% of the #roblem
!
f
"
1
]
1
is arbitrage&free2 +hich in turn is e,uivalent to the other conditions stated in the !rbitrage
Theorem for arket atrices5
These results #oint to a research #rogram that could be develo#ed in several directions5 0ne
direction is to sho+ ho+ the second&order sufficient conditions for o#timalit$ could be inter#reted
economicall$ as the conditions for arbitrage to eliminate its o+n #ossibilit$5 /reliminar$ results in
this direction are outlined in the a##endi%5 !nother direction of develo#ment is to e%tend the
arbitrage inter#retation to other areas of o#timi1ation theor$ such as o#timi1ation +ith ine,ualit$
constraints and o#timal control theor$5
0eferences
Berge2 Claude2 and !5 "houila&.ouri2 <CFD5 (rogramming) *ames and +ransportation
,etorks. Xohn 4e+ 6ork: Wile$ and ons5
Castellan2 "ilbert5 <CF?5 (hysical .hemistry. 4e+ 6ork: !ddison&Wesle$5
Cournot2 !ugustin5 <=C@ 9orig5 <=>=:5 /athematical (rinciples of the +heory of 0ealth Trans5
4athaniel Bloom5 4e+ 6ork: acmillan5
de Finetti2 Bruno5 <CF? 9orig5 <C>@:5 7Foresight1 2ts 3ogical 3as) 2ts 4ub5ective 4ources5 In
4tudies in 4ub5ective (robability ed5 .5 '$burg and .5 mokler2 C>&<D=5 4e+ 6ork:
Xohn Wile$5
<=
Ellerman2 David /5 <C=?5 7!rbitrage Theor$: ! athematical Introduction57 426/ 7evie5 2F:
2?<&F<5
Ellerman2 David /5 <CC05 7!n !rbitrage Inter#retation of Classical 0#timi1ation57
/etroeconomica ?<2 no5 >: 2DC&@F5
"ross2 Xonathan5 <C@?5 7)oltage gra#hs57 8iscrete /ath C: 2>C&?F5
.arar$2 Frank5 <CD>5 70n the notion of balance of a signed gra#h57 /ichigan /ath. 9. 2: <?>&
?F5
.arar$2 Frank2 35 S5 4orman2 and D5 Cart+right5 <CFD5 4tructural /odels5 4e+ 6ork: Xohn
Wile$5
.arar$2 Frank2 B5 *indstrom2 and .5 Setterstrom5 <C=25 70n balance in grou# gra#hs57
,etorks <2: ><@&2<5
'irchhoff2 "5 <=?@5 7Yber die !uflosung der "leichungen2 auf +elche man dei der Lntersuchung
der linearen )erteilung galvanischer trome gefuhrt +ird57 6nnalen der (hysik und
.hemie @2: ?C@&D0=5
odigliani2 Franco2 and erton .5 iller5 <CD=5 7The Cost of Ca#ital2 Cor#oration Finance2
and the Theor$ of Investment57 6merican Economic 7evie ?=: 2F<&C@5
orse2 /hili#5 <CF?5 +hermal (hysics5 4e+ 6ork: W5 !5 BenAamin5
3amse$2 Frank /lum#ton5 <CF0 9orig5 <C2F:5 +ruth and probability5 In +he Foundations of
/athematics) ed5 35 B5 Braith+aite5 /aterson 45X5: *ittlefield2 !dams Z Com#an$5
3oss2 te#hen !5 <C@F5 7The !rbitrage Theor$ of Ca#ital !sset /ricing57 9. Econ. +heory <>:
>?<&F05
3oss2 te#hen !5 <C@F5 73eturn2 3isk and !rbitrage57 In 7isk and 7eturn in Finance2 ed5 Ir+in
Friend and Xames *5 Bicksler5 Cambridge2 ass5: Ballinger5
trang2 "ilbert5 <C=05 3inear 6lgebra and its 6pplications5 econd edition5 4e+ 6ork:
!cademic /ress5
)arian2 .al 35 <C=@5 7The !rbitrage /rinci#le in Financial Economics57 +he 9ournal of
Economic (erspectives <2 no5 2: DD&@25
<C
Appendi#: +econdOrder Conditions as Arbitrage+tabi$ity Conditions
The arbitrage inter#retation e%tends to the second&order sufficient conditions5 There is a natural
stabilit$ notion based on arbitrage5 tabilit$ re,uires that arbitrage eliminate its o+n #ossibilit$5
The e%change of ! for B tends to reduce the rate of e%change of ! into B5 Thus 7transformation
lo+ers the rate of transformation7 is a arshallian conce#t of 7arbitrage stabilit$57 When a##lied
to the market in the mathematics of classical o#timi1ation2 arbitrage stabilit$ is e,uivalent to the
second&order sufficient conditions5
"iven the demand #rice /
D
9,: and su##l$ #rice /

9,: as functions of ,uantit$2 arshallian
stabilit$ re,uires that
dB/
D
9,: ; /

9,:EId, G /
D
( ; /

( [ 05
arshallian stabilit$ is closel$ related to the stabilit$ notion of arbitrage eliminating its o+n
#ossibilit$5 If /
D
9,: V /

9,:2 an arbitrageur could bu$ lo+ at /

9,: and sell high at /
D
9,:5
,
M
/
D

/
Figure 1'. Circular Transformation Lsing u##l$ and Demand /rices
In a circular transformation of mone$ into the commodit$ and back into mone$2 the arbitrageur
nets the amount /
D
9,: ; /

9,:5 Thus arbitrage +ill tend to eliminate its o+n #ossibilit$ if there is
arshallian stabilit$2 /
D
( ; /

( [ 05
Consider a sim#le t+o&variable constrained ma%imi1ation #roblem:
a%imi1e $ G f9%
<
2%
2
:
ubAect to: g9%
<
2%
2
: G b5
!ll derivatives are evaluated at %
o
2 +hich satisfies the first&order necessar$ conditions5 .ence the
market
b $
f Ig
< <
f Ig
2 2
is arbitrage&free2 i5e5 f
<
Ig
<
G f
2
Ig
2
G Q5
Consider an$ differentiable #arameteri1ation %9t: G 9%
<
9t:2%
2
9t:: defined in a neighborhood of %90:
G %
o
that holds b constant2 i5e52
20
dgIdt G g
<
%
<
( K g
2
%
2
( G 0
9+here the #rime denotes differentiation:5 This variation in the % instruments can be thought of as
a circular transformation of $ into b and back into $5 For instance2 if g
2
%
2
( G ;g
<
%
<
( V 02 then R
f
<
%
<
(R units of $ are used u# and transformed into ;9g
<
If
<
:f
<
%
<
( G ;g
<
%
<
( G g
2
%
2
( units of b2 +hich in
turn are transformed into 9f
2
Ig
2
:g
2
%
2
( G f
2
%
2
( units of $5 The net change in $ is
d$Idt G f
<
%
<
( K f
2
%
2
(5
For this circular transformation2 f
<
Ig
<
is the 7su##l$ #rice7 and f
2
Ig
2
is the 7demand #rice7 of b in
terms of the numeraire $5
b $
f Ig
< <
f Ig
2 2
Figure 11. Circular Transformation in 3esource&a%imand arket
Thus 7arshallian stabilit$7 a##lied in this conte%t re,uires that
dBf
2
Ig
2
; f
<
Ig
<
EIdt [ 05
Evaluating the derivative2 dBf
2
Ig
2
; f
<
Ig
<
EIdt G
Bg
2
9f
2<
%
<
( K f
22
%
2
(: ; f
2
9g
2<
%
<
( K g
22
%
2
(:EIg
2
2
; Bg
<
9f
<<
%
<
( K f
<2
%
2
(: ; f
<
9g
<<
%
<
( K g
<2
%
2
(:EIg
<
2
5
ulti#l$ing through b$ the #ositive b G g
2
%
2
( G ;g
<
%
<
( V 0 and using the first&order conditions2
f
<
Ig
<
G f
2
Ig
2
G Q2 $ields
b dBf
2
Ig
2
; f
<
Ig
<
EIdt G
9f
2<
;Qg
2<
:%
<
(%
2
( K 9f
22
;Qg
22
:%
2
(
2
K 9f
<<
;Qg
<<
:%
<
(
2
K 9f
<2
;Qg
<2
:%
<
(%
2
( G
( )
% %
* *
* *
%
%
< 2
<< <2
2< 22
<
2
0
( (
(
(
1
]
1
_
,
<
for an$ 9%
<
(2%
2
(: such that g
<
%
<
( K g
2
%
2
( G 02 +here * is the *agrangian function
*9%
<
2%
2
2Q: G f9%
<
2%
2
: K QBb ; g9%
<
2%
2
:E5
Thus the 7arshallian stabilit$7 conditions ada#ted to this arbitrage conte%t $ield the usual
second&order sufficient conditions for a ma%imum5
The #roblem +ith arshallian stabilit$ conditions is the generali1ation to multi#le markets5 This
#roblem has a sim#le solution +hen reconce#tuali1ed in the arbitrage conte%t5 Instead of vie+ing
f
2
Ig
2
; f
<
Ig
<
as the difference bet+een the 7demand #rice7 and 7su##l$ #rice27 vie+ it as the net
2<
change in $ +hen f
<
Ig
<
units of $ are transformed into 9f
<
Ig
<
:9g
<
If
<
: G < unit of b2 +hich in turn is
transformed into <9f
2
Ig
2
: units of $ for the net change of $ G f
2
Ig
2
; f
<
Ig
<
5
In the circular transformation #reviousl$ described2 Rf
<
%
<
(R units of $ +ere transformed back into
f
2
%
2
( units for the net change d$Idt G f
<
%
<
(K f
2
%
2
(5 ! circular transformation is described b$ an$
variation %9t: +hich leaves b constant2 i5e52 g
<
%
<
( K g
2
%
2
( G 05 ! circular transformation defined in
a neighborhood of a critical #oint %
o
$ields no net change in $2 i5e52 f
i
%
i
( G 05 !rbitrage eliminates
its o+n #ossibilit$ if the rate of change of that net amount is negative5 Thus2 a critical #oint %
o
is
said to have arbitrage stability if for an$ circular transformation defined in a neighborhood of %
o
2
dBf
i
%
i
(EIdt [ 05
The notion of arbitrage stabilit$ e%tends to the 7multi#le markets7 conte%t of the general
o#timi1ation #roblem:
a%imi1e $ G f9%
<
25552%
n
:
ubAect to: g
<
9%
<
25552%
n
: G b
<
W
g
m
9%
<
25552%
n
: G b
m
5
! circular transformation %9t: defined in a neighborhood of a critical #oint %90: G %
o
leaves all the
b
A
constant g
i
A
%
i
( G 0 for A G <25552m5 ince the 7market7 is arbitrage&free2 the circular
transformation $ields no net change in $2 i5e52 f
i
%
i
( G 05 The critical #oint %
o
has arbitrage
stability in this more general conte%t if an$ circular transformation tends to reduce the net change
in $2 i5e52 for an$ circular transformation %9t:2
[ ]
d f %
dt
i i
(
5
< 0
Equation 13. !rbitrage&tabilit$ Condition
This arbitrage&stabilit$ condition is sim#l$ a reformulation of the usual second&order sufficient
condition for a ma%imum2 so +e have an arbitrage inter#retation for the sufficienc$ conditions5
22