Digitized
by
the
Internet
Archive
in
2011
with
funding
from
Boston
Library
Consortium
IVIember
Libraries
HB31
'DEWEV
.M415
'Massachusetts
Institute
of
Technology
Department
of
Economics
Working
Paper
Series
COMPETITION
AND
EFFICIENCY
IN
CONGESTED MARKETS
Daron
Acemoglu
Asuman
Ozdaglar
Working
Paper
05-06
February
16,2005
Revised:
January
20,
2006
Room
E52-251
50
Memorial
Drive
Cambridge,
MA
021
42
This
paper
can be
downloaded
withoutcharge from
theSocial
Science
Research
Network
Paper
CollectionatCompetition and
Efficiency
in
Congested
Markets^
Daron Acemoglu
Department
ofEconomics,
Massachusetts
Institute ofTechnology
Asuman
E.
Ozdaglar
Department
of ElectricalEngineering
and
Computer
Science
Massachusetts
Institute
ofTechnology
January
20,2006
Abstract
We
studythe efSciencyofoligopolyequilibria incongested markets.The
moti-vatingexamplesaretheallocation ofnetworkflows inacommunication networkor
oftrafficina transportation network.
We
showthat increasing competitionamong
oligopoUstscanreduceefficiency, measuredasthedifferencebetweenusers'
will-ingnesstopay anddelaycosts.
We
characterize atightbound
of 5/6onefficiencyin pure strategyequilibria
when
thereis zero latency at zero flow anda tightbound
of2v^
—
2withpositivelatencyatzeroflow. These boundsaretightevenwhen
thenumbersofroutesand ohgopohstsarearbitrarily large.*We
thankXin Huang,RameshJohari,EricMaskin,Eilon Solan, NicolasStierMoses,JeanTirole,JohnTsitsiklis,IvanWerning,
Muhamet
Yildiz, twoanonymousrefereesandparticipants atvarious1
Introduction
We
analyzepricecompetitioninthepresenceofcongestioncosts. Consider thefollowingenvironment: oneunit oftrafficcan use one of/ alternative routes.
More
trafficon
aparticular route causesdelays,exerting anegative (congestion) externality
on
existingtraffic.^ Congestioncosts arecaptured
by
aroute-specificnon-decreasingconvexlatencyfunction, li(•). Profit-maximizingohgopolists set prices(tolls) fortravel
on
each routedenoted
by
Pi.We
analyzesubgame
perfectNash
equilibria ofthisenvironment,where
for eachprice vector,p, alltrafficchooses thepaththathas
minimum
(delayplus toll)cost, li
+
Pi,and
oligopohstschoosepricestomaximize
profits.The
environmentwe
analyze isof practical importance for anumber
ofsettings.These
include transportationand communication
networks,where
additionaluse ofaroute (path) generatesgreatercongestionforall users,
and
markets inwhich
there are"snob"effects,so thatgoods
consumed by
fewer otherconsumers
aremore
valuable(seefor example, [53]).
The
key feature oftheseenvironments is the negative congestionexternalitythatusers exert
on
others. Thisexternalityhasbeen
well-recognized sincethe
work by
Pigou[40] ineconomics,by
[46], [57], [5]intransportationnetworks,and by
[36], [24], [23], [30] in
communication
networks.More
recently,therehasbeen
agrowingliterature that focuses
on
quantification of efficiency loss (referred to as the price ofanarchy) that results
from
externahtiesand
strategic behavior in different classes ofproblems: selfish routing (e.g., [25], [45], [10], [11], [39]
and
[15]); resource allocationby market mechanisms
(e.g., [22], [47], [31], [59]); network design (e.g., [3]);and
two-stagecompetitivefacilitylocationwithout congestioncosts
and
externalities(e.g.,[54]).Nevertheless, the game-theoreticinteractions
between
(multiple) serviceprovidersand
users, orthe effects ofcompetition
among
the providerson
the efficiency losshas notbeen
consideredinnetworks with congestion (externalities). This isan
important areafor analysis since in
most
networks congestion is a first-order issueand
(competing)profit-maximizingentitieschargepricesfor use. Moreover,
we
willshow
that thenatureofthe analysischanges significantly inthepresenceofpricecompetition.
We
provide ageneralframework
for the analysis of price competitionamong
ser-vice providers^inacongested(andpotentiallycapacitated)network,studyexistence of
purestrategy
and
mixed
strategy equihbria,and
characterizeand
quantifytheefficiencyproperties ofequilibria.
There
are foursetsofmajor
resultsfrom ouranalysis.First,
though
the equilibrium of traffic assignment without prices can be highlyinefficient (e.g., [40], [45], [10]), price-settingby amonopolist internalizesthe negative
externality
and
achievesefficiency.Second,increasingcompetitioncanincreaseinefficiency. Infact,changingthe
market
structure
from
monopoly
toduopoly
almost always increases inefficiency. This resultcontrastswith
most
existing resultsintheeconomicsliteraturewhere
greatercompetitiontendstoimprovethe allocation of resources(e.g. see Tirole [51]).
The
intuitionfor thisresult,
which
isrelated to congestion,is illustratedby
theexample
we
discuss below.^^
An
externalityariseswhenthe actionsofthe playerinagameaffectsthe payoff ofotherplayers.^We
useoligopolistandservice providerinterchangeablythroughoutthe paper.^Because,inourmodel,users arehomogeneous and have aconstant reservationutility,intheabsence
Third
and most
important,we
providetightbounds
on theextent ofinefficiencyinthepresenceofohgopohsticcompetition.
We
show
thatwhen
latency at zero flow(traffic)is equal tozero, socialsurplus (defined as the difference
between
users' willingness topay
and
the delay cost) inany
pure strategy oligopoly equilibriumis always greaterthan 5/6 ofthe
maximum
socialsurplus.When
latency at zero flow canbe positive,there isaslightlylower
bound
of 2\f2-
2«
0.828. Thesebounds
areindependent ofboththe
number
of routes, /, which couldbe arbitrarily large,and
how
these routesare distributed acrossdifferentoligopolists (i.e., ofmarketstructure). Simpleexamples
reachthesebounds.
Finally,
we
alsoshow
that purestrategy equilibriamay
fail to exist. This is notsurprisinginviewofthefactthat
what we
havehereisaversion ofaBertrand-Edgeworthgame
where
pure strategy equilibria do not exist in the presence of convex costs ofproductionorcapacity constraints (e.g., [14], [49], [7], [56]). However,in ourohgopoly
environment
when
latency functions arelinear,a purestrategyequilibriumalwaysexists,essentiallybecause congestionexternalities
remove
the payoff discontinuities inherentintheBertrand-Edgeworth game. Non-existence
becomes
anissuewhen
latency functionsare highly convex. In this case,
we
provethatmixed
strategy equilibria always exist.We
alsoshow
thatmixed
strategyequihbriacanlead to arbitrarilyinefficientworst-caserealizations; in particular, social surplus can
become
arbitrarily smallrelative to themaximum
socialsurplus,though
the average performanceofmixed
strategyequihbriais
much
better.The
followingexample
illustratessome
ofour results.Example
1 Figure 1shows
asituation similar to theonefirst analyzedby
Pigou[40]tohighlighttheinefficiency
due
tocongestionexternalities.One
unit oftrafficwilltravelfrom origin
A
todestinationB, using either route 1or route2.The
latency functionsare given
by
x^ 2
'i(a;)
=
y
,
k
(x)=
-X.Itisstraightforward to see that theefficientallocation [i.e.,onethatminimizesthe total
delay cost
^^hix^Xi^
isxf
=
2/3and
xf—
1/3, while the (Wardrop) equilibriumallocationthatequates delay
on
thetwo
pathsisx^^
k,.73>
x\ and
x^^
f« .27<
xf
.
The
sourceofthe inefficiencyisthateachunit oftraffic does notinternalizethe greaterincrease indelayfromtravel
on
route 1,sothere istoomuch
use ofthisrouterelativetotheefficientallocation.
Now
consider a monopolist controlhng both routesand
setting prices for travel tomaximize
itsprofits.We
show
below that inthis case, themonopolist will setapriceincludinga
markup,
Xil\(when
li isdiflFerentiable), whichexactly internalizes thecon-gestionexternality. Inotherwords,this
markup
isequivalent to thePigovian taxthatasocialplanner
would
setinorder toinducedecentralizedtraffictochoosetheefficiental-location. Consequently,in thissimpleexample,
monopoly
priceswillbe pf^^=
(2/3) -\rkand
p^^
=
(2/3^)-I-/c, forsome
constantk.The
resultingtraffic intheWardrop
equi-hbrium
will beidenticaltothe efficientallocation, i.e.,x'^^=
2/3and
x^^
=
1/3.l,(x)=x /3
1 unitof traffic
Figure 1:
A
two
link network with congestion-dependantlatency functions.Finally, consider a
duopoly
situation,where
each routeiscontrolledby
a differentprofit-maximizingprovider. Inthiscase,itcanbe
shown
thatequilibriumpriceswilltaketheform
pf^
—
Xj{l[+
/j)[seeEq.(20)inSection4],ormore
specifically,pf
^^
0.61and
p^^
^
0.44.The
resulting equilibrium trafficisx^^
ft; .58<
xf and
x^^
«
.42>
xf,which
alsodiffersfrom
theefficientallocation.We
willshow
thatthisisgenerally the casein the ohgopolyequilibrium. Interestingly, while in the
Wardrop
equilibrium withoutprices, there
was
toomuch
trafficon
route 1,now
there is too little traffic becauseof its greater
markup.
It is also noteworthy that although theduopoly
equilibriumis inefficient relative to the
monopoly
equilibrium, in themonopoly
equilibrium k ischosen suchthatallofthe
consumer
surplusiscapturedby
themonopolist,while in theoligopolyequihbriumusers
may
havepositiveconsumer
surplus.^The
intuition for the inefficiency ofduopoly
relative tomonopoly
is related to anew
source of(differential)monopoly power
for eachduopolist,which
theyexploitby
distortingthepatternoftraffic;
when
provider 1, controllingroute 1,charges a higherprice,itrealizesthatthis will
push
some
trafficfromroute1toroute2,raisingcongestionon
route2.But
thismakes
thetrafficusing route 1become more
"locked-in," becausetheiroutside option, travel
on
theroute2,hasbecome
worse.^As
aresult,theoptimalpricethateach duopolistcharges will include
an
additionalmarkup
over the Pigovianmarkup.
These
are Xil'2 for route 1and
xj^j for route 2. Since thesetwo markups
are generally different, they will distort the pattern oftraffic
away from
the efficientallocation. Naturally, however, prices are typicallylowerwith duopoly,so even
though
socialsurplus declines, userswillbebetteroffthanin
monopoly
(i.e.,theywillcommand
apositive
consumer
surplus).There
isalargehteratureon
modelsofcongestionboth
intransportationand
commu-nicationnetworks (e.g. [5], [38], [44], [33], [34], [45]).^ However, very fewstudieshave
^Consumersurplusisthe differencebetweenusers'willingness topay(reservationprice)andeffective
costs, Pi
+
Zi(Xj), and is thus different from social surplus (which is the difference betweenusers'willingness topay andlatencycost, li{xi),thusalsotakesintoaccount producersurplus/profits). See
[32].
^Using economicsterminology,wecouldalsosay that thedemandforroute1becomesmore
"inelas-tic". Sincethistermhaisadifferent meaninginthecommunication networksliterature (see[48]),we
do notuseithere.
®Someofthesepapersalsouse prices(or tolls)toinduce flow patterns that optimizeoverallsystem
investigated the implications ofhaving the "property rights" over routes assigned to
profit-maximizingproviders. In [4],Basar
and
Srikantanalyzemonopoly
pricingunderspecific assumptions on the utihty
and
latency functions.He
and
Walrand
[19] studycompetition
and
cooperationamong
internet service providersunder specificdemand
models. Issues ofefficientallocation of flows ortrafficacrossroutes
do
notariseinthesepapers.
Our
previouswork
[1] studiesthemonopoly problem and
containstheefficiencyofthe
monopoly
result, butnone
ofthe other results here.More
recent independentwork by
[3] buildson
[1]and
alsostudiescompetitionamong
serviceproviders. Usingadifferentmathematical approach, they providenon-tight
bounds on
theefficiency lossforthe case ofelastictraffic. Finally, incurrentwork, [2],
we
extendsome
ofthe resultsofthispapertoanetwork withparallel-serialstructure.
In the rest of the paper,
we
use the terminology of a (communication) network,though
all of the analysis applies to resource allocation in transportation networks,electricity markets,
and
other economicapplications. Section 2 describes the basicen-vironment. Section 3 brieflycharacterizesthe
monopoly
equilibriumand
establishesitsefficiency. Section 4 defines
and
characterizesthe oligopoly equilibriawithcompeting
profit-maximizing providers. Section 5containsthe
main
resultsand
characterizes theefficiency properties ofthe oligopoly equihbrium
and
providebounds on
efficiency.Sec-tion 6 providesatightefficiency
bound
when
theremay
bepositivelatencyatzeroflow.Section 7 containsconcluding
comments.
Regardingnotation,allvectors areviewedas
column
vectors,and
inequalitiesare tobeinterpretedcomponentwise.
We
denoteby
Mf^theset ofnonnegative /-dimensionalvectors. LetCibe aclosedsubset of[0,oo)
and
let/: Cjh->R
beaconvexfunction.We
use df{x) todenotethesetofsubgradientsof
/
atx,and
f~{x)and
f'^{x) todenotetheleft
and
rightderivatives of/at x.2
Model
We
consider anetwork with/parallel links. LetI =
{1,...,/} denotethesetoflinks.
LetXidenotethe total flow
on
linki,and
x=
[xi,...,xj]denotethe vector oflink flows.Each
link inthenetwork has a fiow-dependentlatency function /i(x,),which measuresthe traveltime (ordelay) asa function ofthetotalflow
on
linki.We
denotethe priceperunit flow(bandwidth)oflinki
by
Pi. Letp
=
[pi,...,p/]denotethe vector ofprices.We
are interestedintheproblem
ofroutingdunits offlow acrossthe /links.We
as-sume
thatthisisthe aggregateflow ofmany
"small" usersand
thusadopttheWardrop'sprinciple (see [57]) incharacterizingtheflow distributioninthenetwork;i.e.,the flows
arerouted along paths with
minimum
effective cost, defined as thesum
ofthelatencyatthe given flow
and
the price of that path (see the definition below).^We
alsoas-sume
that the usershave areservationutilityR
and
decidenot tosendtheirflowiftheeffectivecost exceedsthe reservationutility. This impliesthat user preferencescan be
induce optimalflows,withthe goal of choosingtollsfromthissetaccording tosecondarycriteria,e.g.,
minimizingthetotalamountoftollsor thenumberoftolledroutes; see[8],[21],[28],[27],and[20].
^Wardrop's principleisused extensivelyinmodellingtraffic behaviorintransportation networks,
u(x)
Figure2: Aggregateutilityfunction.
represented
by
thepiecewise linearaggregateutilityfunctionu
(•)depictedinFigure2.^To
accountforadditional side constraintsinthetrafficequihbrium problem,includingcapacityconstraints
on
thelinks,we
use thefollowing definition ofaWE
(see[29], [26]).Lemma
1shows
that this definition is equivalent tothemore
standard definition of aWE
usedinthe literatureundersome
assumptions.Definition
1 For a given price vectorp
>
0,^equilibrium
(WE)
ifa vector
x
G
argmax
<.2_]{R~ki
xY'^)-p,)x,
is a
Wardrop
(1)
We
denotetheset ofWE
at a givenp by W{p).
Assumption
1 For eachi£
Z, the latencyfunction kconvex, nondecreasing,
and
satisfies /,(0)—
0.[0,oo) 1-^ [0,do] isclosed,^°
The
assumptionofzerolatencyatzeroflow, i.e.,/j(0)=
0,imphes
thatalllatencyisdue
to flow oftraffic,and
there arenofixedlatencycosts.^^ It isadoptedto simplify thediscussion, especiallythe characterization ofequilibriumpricesin Proposition9 below.
A
trivialrelaxation ofthis assumption to ^,(0)=
L
for all ie
2
forsome
L
>
willhave
no
effecton
any
ofthe results in the paper. Allowing for differential levels of^Thissimplifyingassumptionimpliesthatallusers are "homogeneous" inthe sense thattheyhave
thesamereservationutility,R. Theanalysisbelowwillshowthat the value ofthisreservationutility
R
hasnoeffecton anyoftheresultsaslong asitisstrictly positive.We
discuss potentialissues inextendingthisworktouserswithelasticand heterogeneousrequirementsintheconcludingsection.
^Since the reservationutilityof users
isequal toR,wecanalso restrictattention topi<
R
foralli. Throughoutthepaper,weusep> and p £[0,
RY
interchangeably.'''Forafunction /:R"i—»(—00,00],wesay that /isclosedifthelevel set{x\f{x
<
c)}isclosedforevery scalarc. NotethatafunctionisclosedifandonlyifitislowersemicontinuousoverR"(see
[9],Proposition1.2.2).
''Thisassumption would be agood approximationtocommunication networks where queueingdelays
/j(0) complicates the analysis, but has littleeffect onthe
major
results. This caseisdiscussedinSection6,
where
we
provideaslightlylower tightbound
forthe inefficiencyofohgopoly equihbria withoutthisassumption.
Another
feature ofAssumption
1 isthatit allowslatency functions tobe extendedreal-valued, thus allowingfor capacityconstraints. Let Ci
—
{xE
[0,oo) | li{x)<
oo}denote the effective
domain
of/,.By
Assumption
1, Cjis a closed interval of theform
[0, b]or[0,oo). Let
bd
—
sup^^,^, x.Without
lossofgenerahty,we
canadd
the constraintXi
6
Ciin Eq. (1). Usingthe optimality conditionsforproblem
(1),we
seethata vector^WE
£i^J^ isa
WE
ifand
onlyifJ^iei
^T^ —
^^'^'^thereexistssome
A>
suchthatK
EiGi
^^^
~d)
=
and
foralli,R
-
kixf"^)-Pi
<
A ifxf"^-
0, (2)=
AiiO<xf'^
<bc„
>A
ifxr^
=
6c,.When
the latency functions are real-valued[i.e.,Cj=
[0,oo)],we
obtainthe followingcharacterization ofa
WE,
which
isoftenusedasthedefinition ofaWE
intheliterature.This
lemma
states that in theWE,
the effective costs, defined as li{xf^^)+
pi, areequalized
on
alllinkswithpositive flows.Lemma
1 LetAssumption
1 hold,and assume
further that Ci—
[0,oo) forall iG
T.Then
anonnegativevector x*£
W{p)
ifand
onlyifli{x*)
+
Pi—
mm{lj{x*)+
pj},V
iwithX*>
0, (3)j
kix*)
+
Pi<
R,V
z withX*>
0,iei
withX^jgjS;*
=
dliminj{lj{xj)+
Pj}<
R-Example
2 belowshows
that condition (3) in thislemma
may
not holdwhen
thelatency functions arenotreal-valued.
The
existence, uniqueness,and
continuityprop-ertiesofa
WE
arewell-studied (see[5], [12], [50]).We
provideherethestandard proofforexistence,basedonestablishingtheequivalence of
WE
and
theoptimalsolutions ofaconvexnetwork optimization problem, which
we
willrefertolater inour analysis.Proposition
1(Existence
and
Continuity)
LetAssumption
1hold. Forany
pricevector
p
>
0, the set ofWE,
W{p),
is nonempty. Moreover, the correspondenceW
:M^
^
R^
isupper semicontinuous.Proof.
Given any
p>
0,consider the followingoptimizationproblem
maximize^>o
^
\iR
-
Pi)xi-
j li{z)dz\ (4)subject to 2_.^i
^
d.Inviewof
Assumption
(1)(i.e.,liisnondecreasingforall i), itcan beshown
that theobjective function of
problem
(4) is convexoverthe constraintset,which
isnonempty
(since
G
Ci)and
convex. Moreover, thefirstorder optimalityconditions ofproblem
(4),whichare alsosufficientconditionsforoptimality,are identical to the
WE
optimalityconditions [cf. Eq. (2)].
Hence
a flow vectorx^^
E
W{p)
ifand
onlyifitis an optimalsolution of
problem
(4). Since the objective function ofproblem
(4) is continuousand
theconstraintsetiscompact,this
problem
hasan optimalsolution,showingthatW{p)
isnonempty.
The
fact thatVF
isan upper
semicontinuous correspondence at everyp
follows
by
usingtheTheorem
oftheMaximum
(see Berge [6], chapter 6) forproblem
(4).
Q.E.D.
WE
flows alsosatisfyintuitivemonotonicitypropertiesgiveninthe followingpropo-sition.
The
prooffollowsfrom
the optimalityconditions [cf.Eq. (2)]and
isomitted(see
[I])-Proposition
2(Monotonicity)
LetAssumption
1 hold. For a givenp
>
0, letP-j
=
\PiWj-(a) For
some p
<
p, letx£
M^(p)and x E W{p).
Then,J^iei^^—
Yliei^i-(b) For
some
pj<
pj, letx E W{pj,p^j) and x E
W{pj,p-j).Then
Xj>
Xjand
Xi<
Xi,foralli
^
j.(c) For
some
X
C
I,supposethatpj<
pj foralljE
I
and
pj=
pjfor allj^
X,and
letX
eW{P)
and
x E W{p).
Then
Yljei^j—
X^ief^j-For agiven price vectorp, the
WE
need not be unique in general.The
followingexample
illustratessome
properties oftheWE.
Example
2 Consider atwo
linknetwork. Letthe total flowhed=
1and
thereserva-tionutility
be
i?=
1.Assume
that the latency functions are givenby
liix)
—
MX]
=
< , . 3^
'^
'\ oo otherwise.
At
the price vector (pi,P2)—
(1,1), the set ofWE,
W{p),
is givenby
the set of ahvectors (xi,2;2) with
<
x^<
2/3and
X^^Xj<
1.At any
price vector (pi,P2) withPi
>
P2=
1,W{p)
isgivenby
all (0,X2) with<
X2<
2/3.This
example
also illustratesthatLemma
1 need not holdwhen
latencyfunctionsare not real-valued. Consider, for instance, the price vector (pi,P2)
=
(1—
e,1—
ae)for
some
scalara>
1. In thiscase,theuniqueWE
is(xi,X2)=
(1/3,2/3),and
clearlyeffective costs
on
thetwo
routes are notequalized despitethefactthattheyboth
havepositive flows. This arises because the path with the lower effective cost is capacity
constrained, sono
more
trafficcanuse that path.Under
further restrictionson
the/,,the followingstandardresultfollows(proofProposition
3(Uniqueness)
LetAssumption
1 hold.Assume
further that k isstrictly increasingover C^. For
any
price vectorp>
0, the set ofWE,
W{p),
is asingleton. Moreover,the function
W
:R^
i—>M^
iscontinuous.Since
we
do
notassume
thatthelatency functions are strictlyincreasing,we
needthe following
lemma
inour analysis to dealwithnonuniqueWE
flows.Lemma
2 LetAssumption
1hold. Fora givenp
>
0,definethesetJ={i€l|3x,
X e
W{p)
withXij^Xi}. (5)Then
kixi)
=
0,V
ie
J,V
x G Wip),
Pi
=
Pj,V
i, jei.
Proof.
Considersome
iG
2"and
x € W{p).
Sincei 6X, thereexistssome x 6
W{p)
suchthatXi
^
Xi-Assume
withoutlossofgeneralitythat x,>
Xj. Therearetwo
cases to consider:(a) If Xk
>
Xk for all k j^ i, then X^jsi^j>
X^iei-^J' whichimphes
that theWE
optimality conditions[cf. Eq. (2)] forx hold with A
=
0.By
Eq. (2)and
Xj>
x,,we
haveli{xi)
+Pz<
R,li{xi)+Pi>R,
which
togetherimply thatli{xi)=
li{xi).By
Assumption
1 (i.e., kis convexand
^.(0)
=
0),it followsthatli{xi)=
0.(b) IfXk
<
Xk forsome
k,by
theWE
optimahtyconditions,we
obtaink{Xi)
+
Pi<
IkiXk)+
Pk,k{ii)
+Pi>
lk{xk)+Pk-Combining
the above withXi>
Xiand
x^<
x^,we
see that li{xi)=
k{xi),and
hi^k)
=
hi^k)-By
Assumption
1, this shows that li{Xi)—
(and also thatPi
=Pk)-Next
considersome
z, jG
2".We
willshow
thatpi
=
pj. Since iE
I, thereexistX, XE.
W{p)
suchthatXt>
Xi.There
arethree cases to consider:• Xj
>
Xj. IfXk>
Xk for all k^
i,j, thenX^^Xm <
d, implying thattheWE
optimalityconditionsholdwithA
=
0. Therefore,we
havek{xi)
+
Pi<
R,lj{xj)
+
pj>
R,which
togetherwithli{xi)=
lj{xj)=
imply thatPi=
Pj.• Xj
=
Xj. Since j€
X,by
definition theremust
existsome
otherx^W{p)
suchthatXj
^
Xj. Repeating theabovetwo
stepswith Xj instead of Xj yields the desiredresult.
Q.E.D.
Intuitively, this
lemma
states that ifthere exist multipleWEs,
x,x E
W{p)
suchthatXi
^
Xi,thenthe latency function/,must
belocallyflataround
Xi (andXi).Given
the assumptionthat/j(0)
=
and
the convexityoflatency functions, thisimmediatelyimpliesli{xi)
=
0.We
nextdefinethesocialproblem
and
the socialoptimum, which
istherouting(flowallocation) that
would
be chosenby
aplanner that hasfullinformationand
fullcontroloverthenetwork.
Definition 2
A
flow vectorx^
is asocialoptimum
ifit isan
optimalsolution of thesocialproblem
m£iximizej:>o
y^iR
—
li{xi)jXi (6)iei
subject to
y.
^i^
d-Inviewof
Assumption
1,thesocialproblem
has a continuousobjectivefunctionand
a
compact
constraintset,guaranteeingthe existence ofasocialoptimum,
x^. Moreover,usingthe
optimaUty
conditionsfora convexprogram
(see [9],Section4.7),we
seethata vector
x^ E
R^
is asocialoptimum
ifand
only if^^^jxf
<
dand
there existsasubgradient gi.
e
dli{xf) foreachz,and
aA'^>
suchthat A'^(^jgjxf
—
d)—
Qand
for each2,
R
-
k{xf)-
xfgi^<
A^ ifxf
=
0, (7)=
A^
ifO<xf<bc„
>A^
iixf=
bcrForfuture reference, fora given vectorx
€
Mf^,we
definethe value of the objectivefunctioninthe socialproblem,
S{x)
=
Y,iR-l^[x^))Xi,
(8)asthe socialsurplus,i.e., the difference
between
users' willingnesstopay and
the total3
Monopoly
Equilibrium
and
Efficiency
In this section,
we assume
that amonopoUst
service providerowns
the / linksand
chargesaprice ofpiper unit
bandwidth on hnk
i.We
considered arelatedproblem
in[1] foratomicusers withinelastic traffic (i.e.,theutilityfunction ofeachofa finite set
ofusers is a step function),
and
with increasing,real-valuedand
differentiablelatencyfunctions. Here
we
show
that similarresultsholdforthemore
generallatency functionsand
thedemand
model
considered inSection2.The
monopolist setsthe prices tomaximize
his profitgivenby
n(jD,x)
^^p,Xi,
where x €
W{p). Thisdefinesatwo-stagedynamic
pricing-congestiongame,where
themonopolist sets pricesanticipating the
demand
ofusers,and
given the prices (i.e., ineachsubgame),userschoosetheirflow vectorsaccordingto the
WE.
Definition 3
A
vector(p^^,x^^)
>
is aMonopoly
Equilibrium(ME)
ifx^^
G
W{p^'^) and
liip"'^^,x'^^)
>
n(p,x),V
p
>
0,V
X GW
(p).
Our
definition oftheME
isstronger thanthestandardsubgame
perfectNash
equi-libriumconceptfor
dynamic
games.With
aslightabuseof terminology,letus associatea
subgame
perfectNash
equilibrium with the on-the-equilibrium-path actions of thetwo-stagegame.
Definition
4
A
vector (p*,x*)>
isasubgame
perfect equihbrium(SPE)
ofthepricing-congestion
game
ifx*£ W{p*) and
for allp
>
0, there exists x EW
(p) suchthat
n(p*,x*)
>
n(p,.T).The
followingpropositionshows
thatunderAssumption
1,thetwo
solutionconceptscoincide. Sincethe proofis not relevantfor therest oftheargument,
we
provide itinAppendix
A.Proposition 4
LetAssumption
1 hold.A
vector {p'^'^^,x'^^) isanME
ifand
onlyifitisan
SPE
of the pricing-congestiongame.Sincean
ME
(p*,x*) isan optimalsolution ofthe optimizationproblem
maximizep>o,x>o
^^Pi^i
(9)iei
subject to x
E W{p),
itis easierto
work
withthan anSPE.
Therefore,we
useME
asthe solutionconceptinthispaper.
The
precedingproblem
has an optimalsolution,which
establishes the existenceofan
ME.
Moreover,we
have:Proposition
5 LetAssumption
1hold.A
vectorxisthieflowvectoratanME
ifand
onlyifit isa social
optimum.
Moreover, if(p,x) isan
ME,
thenfor all iwithXi>
0,we
havepi= R
—
k{xi).This proposition therefore establishes that the flow allocation at an
ME
and
thesocial
optimum
are the same. Itsproofis similar toan
analogousresult in [1]and
isomitted.
In addition to the socialsurplus defined above, it is also useful to definethe
con-sumer
surplus,asthedifferencebetween
users'wilhngnesstopay and
effectivecost,i.e.,^j^^ {R
—
k{xi)—pi)xi(see[32]).By
Proposition5, it isclearthat eventhough
theME
achievesthesocial
optimum,
allofthe surplusiscapturedby
themonopohst, and
usersare just indifferent
between
sendingtheirinformationornot (i.e.,receiveno
consumer
surplus).
Our major
motivation for the study of oligopolistic settings isthat they providea betterapproximation to reality,
where
there istypically competitionamong
serviceproviders.
A
secondary motivationisto seewhether anoligopolyequilibriumwillachieveanefficient allocationlikethe
ME,
whilealsotransferringsome
orallofthe surplus totheconsumers.
4
Oligopoly
Equilibrium
We
suppose thatthere areS
serviceproviders, denote the set of serviceprovidersby
«S,
and assume
that eachserviceprovidersE
S
owns
a different subset X, of thelinks.Serviceproviderscharges apricepiperunit
bandwidth on
linki EIs-Given
thevectorof prices of links
owned by
other service providers,p_s=
[Pili^is^ ^^^ profit of serviceprovidersis
^s{Ps,P-s,x)
^
^PiXi,
ieis
for X
e
W{ps,p-s),where
Ps=
\pi]ieis-The
objective ofeachserviceprovider, like themonopolist in theprevioussection,is to
maximize
profits. Becausetheir profitsdepend on
theprices setby
other serviceproviders, each service provider forms conjectures about the actions ofother service
providers, as well asthebehaviorofusers, which,
we
assume, they do accordingtothenotion of
(subgame
perfect)Nash
equilibrium.We
refer to thegame
among
serviceproviders asthe price competitiongame.
Definition 5
A
vector [p'-'^,x^^)>
isa(pure strategy) OligopolyEquilibrium(OE)
if
x°^
€
W
(p°^,p?f
)and
forallseS,
n,(p?^,pef,xO^)
>
n,(p3,pef,x),vp,
>
0,vx
G
w{p,,p'^f). (lo)We
refertop°^
astheOE
price.As
forthemonopoly
case,there isa close relationbetween
apurestrategyOE
and
a pure strategy
subgame
perfect equiUbrium.Again
associating thesubgame
perfectequilibriumwith the on-the-equilibrium-pathactions,
we
have:Definition 6
A
vector {p*,x*)>
isasubgame
perfectequilibrium(SPE)
ofthe pricecompetition
game
ifx* €W
(p*)and
there exists afunctionx
:R^
i->R^
such thatx{p)
e
W
(p) for allp
>
and
foralls eS,n,(p:,pl„a;*)>n,(p„pl„x(p3,plj)
Vp, >0.
(11)The
followingproposition generalizesProposition4and
enables us towork
withtheOE
definition,whichismore
convenientforthesubsequentanalysis.The
proofparallelsthat ofProposition4
and
isomitted.Proposition
6 LetAssumption
1hold.A
vector {p'-'^,x^^)isan
OE
ifand
onlyif itis
an
SPE
ofthe pricecompetitiongame.The
pricecompetitiongame
isneitherconcave nor supermodular. Therefore,classicalargumentsthat areusedto
show
the existence of apurestrategyequilibriumdo not hold(see[16], [52]). Inthenextproposition,
we
show
thatfor linearlatency functions, thereexistsapurestrategy
OE.
The
proofisprovidedintheappendix.Proposition
7 LetAssumption
1hold,and assume
furtherthat the latency functionsarelinear.
Then
the pricecompetitiongame
hasapurestrategyOE.
The
existence result cannot be generalized to piecewise linear latency functions ortolatency functions
which
are linear over theireffective domain, as illustratedin thefollowingexample.
Example
3 Consideratwo
linknetwork. Letthe total flowhe d=
1.Assume
that thelatency functions are given
by
I I \ n 1 f \
Jo
if0<x<5
for
some
e>
and
5>
1/2,withtheconventionthatwhen
e=
0, l2{x)=
ooioi x>
5.We
firstshow
that thereexists no purestrategy oligopoly equilibrium forsmall e(i.e.,thereexists
no
purestrategysubgame
perfectequihbrium).The
followinglistconsidersallcandidateohgopolyprice equilibria (pi,P2)
and
profitable unilateral deviations foresufficientlysmall, thusestablishingthenonexistenceofan
OE:
1. Pi
=
P2=
0:A
small increase in the price of provider 1 will generate positiveprofits,thusprovider 1has an incentive to deviate.
2. Pi
—
P2>
0: Letx
bethe flow allocation at theOE.
IfXi=
1,thenprovider 2-hasanincentive to decreaseitsprice. Ifxj
<
1,thenprovider 1hasan
incentivetodecreaseitsprice.
3.
<
pi<
P2: Player 1has anincentive to increaseitspricesinceitsflow allocationremainsthesame.
4.
<
P2<
Pi'- For esufficientlysmall, theprofit function of player 2, givenpi, isstrictlyincreasing asafunction ofp2, showingthat provider 2 has
an
incentive toincreaseitsprice.
We
nextshow
thatamixed
strategyOE
alwaysexists.We
defineamixed
strategyOE
asamixed
strategysubgame
perfectequilibriumofthe pricecompetitiongame
(seeDasgupta and
Maskin, [13]). Let5"
be the spaceofall (Borel) probability measureson [0,/?]". Let Is denote the cardinality ofT^, i.e., the
number
of links controlledby
serviceproviders. Let/i^
€
B'^be aprobabilitymeasure,and
denotethe vectoroftheseprobabilitymeasures
by
/j,and
the vector of these probabilitymeasures excluding sby
Definition 7 {/j,*,x*{p))isa
mixed
strategy OligopolyEquilibrium(OE)
ifthefunctionx*{p)
G
W
(p) foreveryp G
[0,RY
and
/ Us{ps,p-s,x*{ps,p-s)) d{iil{ps)
X
fl*_^{p_s))JlO,R]'
>
/ Us{Ps,P-s,X*{Ps,P-s))d{fls{Ps) XP'ls(P-s))JlO,R\'
foralls
and
fig€ B^\
Therefore,a
mixed
strategyOE
simplyrequiresthat therebe noprofitable deviationtoadifferentprobability
measure
foreach ohgopoUst.Example
3(continued)
We
now
show
thatthe following strategyprofile is theunique
mixed
strategyOE
for theabovegame when
e—
> (amixed
strategyOE
alsoexists
when
e>
0,butitsstructureismore comphcated and
lessinformative):(
0<P<R{l-S),
fxi{p)=l
1-M«
R{l-S)<p<R,
{ 1 otherwise, (0<P<R{l-S),
M2(P)=<
1-^
R{l-S)<p<R,
I 1 otherwise.otherwise.Notice that ^i has
an
atom
equalto 1—
5 ati?.To
verifythat this profileis amixed
strategy
OE,
let n'be
thedensity offj.,with the conventionthatfj,'=
oowhen
thereisan
atom
atthatpoint. LetMi
—
{p \ji'^(p)
>
0}.To
establish that {111,^2)isamixed
strategy equilibrium, itsufficesto
show
thatthe expectedpayoff to playeriisconstantfor all Pi
G
Mi when
the other player choosesp_i accordingto ;U_j (see [37]). Theseexpectedpayoffs are
n
{pi IiJL-i)=
/ Iii{pi,p-i,x {pi,p-i))dii-i(p_i)
.
(12)
The
WE
demand
x{pi,p2) takes the simple form of xi {pi,P2)=
1 if Pi<
P2and
xi {pi,P2)
=
l—
5iipi>
p2-The
exactvalue of xi (pi,P2)=
1whenpi
=
p2isimmaterialsincethisevent
happens
withzero probability. It isevidentthat the expressionin (12)is constant for allp,
€
Mj
fori=
1,2 given /^iand
112 above. This establishes that{111,112) is a
mixed
strategyOE.
Itcan alsobeverifiedthat there areno
othermixed
strategyequilibria.
The
next proposition,which
is proved inAppendix
B, estabUshes that amixed
strategy equilibriumalwaysexists.
Proposition
8 LetAssumption
1hold.Then
the pricecompetitiongame
hasamixed
strategy
OE,
(/x°^,x°^(p)).We
next providean explicitcharacterization ofpure strategyOE.
Though
ofalsoindependentintei'est, these results are
most
usefulforus to quantify theefficiency lossofoligopolyinthenextsection.
The
followinglemma
shows that an equivalent toLemma
1 (which requiredreal-valued latency functions) also holds with
more
general latency functions at the purestrategy
OE.
Lemma
3 LetAssumption
1hold. If(79°^,x"^^) isapurestrategyOE,
then/,(x°^)+pf^
=
min{L(xf^)
+
pf^},Vzwithxf^>0,
(13)/i(xf^)+pf^
<
R,Viwithxf^>0,
(14)^xp^
<
d, (15)with E,:ei^z°^
=
^ifminj{/j(x^°^)+
Pj}<
R.Proof. Let (p°^,x'^^) bean
OE.
Since x°-^G W{p°^),
conditions (14)and
(15) followbythe definition ofa
WE.
Considercondition(13).Assume
that thereexistsome
i,jG
I
with
xf
^>
0,x°^
>
suchthatUsingtheoptimahtyconditionsfora
WE
[cf.Eq. (2)],thisimpliesthatxf^
=
be,-Con-siderchanging
pf^
topf
^+
eforsome
e>
0.By
checkingthe optimality conditions,we
seethat
we
can chooseesufficientlysmallsuchthatx°^
G
W{pf^
+
e,p^f).Hence
theserviceprovider that
owns
linkicandeviate topf
^
+
eand
increaseitsprofits,contra-dicting thefactthat (p'^^,x'^^) isan
OE.
Finally,assume
to arrive at acontradictionthatminj{/j(x?^)4-Pj}
<
R
and Yliei^?^
<
d. Usingthe optimality conditionsforaWE
[Eq. (2)withA
=
since^j^ixf^
<
d],thisimpliesthatwe
must
havexf^
—
bd
for
some
i.With
asimilarargument
to above, a deviation topf^
+
e keepsx'-'^ asaWE,
and
ismore
profitable,completingthe proof.Q.E.D.
We
needthe following additionalassumptionforourpricecharacterization.Assumption
2Given
apurestrategyOE
(p*^^,x^^),ifforsome
ie
I
withxf^ >
0,we
haveli{xf^)—
0, thenX^—
{i}.Note
thatthisassumptionisautomaticallysatisfiedifalllatency functions arestrictlyincreasingorifallserviceproviders
own
onlyonelink.Lemma
4
Let (p°^,a;°^) be a purestrategyOE.
LetAssumptions
1and
2 hold. LetHa
denotethe profit of serviceprovider sat (jP^,x'~'^).(a) IfUs'
>
forsome
s'6
5,thenfls>
forallsE
S.(b) If
n,
>
forsome
se
<S,thenpf
^x°^
>
foralljE
Is-Proof.
(a) For
some
j£
Jy,defineK
=
p'^^+
lj[x^^),whichispositive since Hgi>
0.Assume
Hg
=
forsome
s. ForkE
Xg, considerthe price pk=
K
—
e>
iorsome
smalle
>
0. Itcan be seenthat attheprice vector {pk,p?.k), thecorrespondingWE
linkflow
would
satisfy Xk>
0. Hence,serviceprovider shasanincentive todeviate topk at
which
hewillmake
positiveprofit,contradictingthe factthat {p'~'^,x^^) isa purestrategy
OE.
(b) Since lis
>
0,we
havep^x^
>
forsome
m
E
X^.By
Assumption
2,we
canassume
withoutlossofgeneralitythat Imix^^)>
(otherwise,we
aredone). Let jE
Xgand
assume
to arrive at acontradiction thatpf^xf^
=
0.The
profit ofserviceprovider satthe purestrategy
OE
can bewritten asn^
=
n^
+
p,PE OE
where
Da denotestheprofitsfromlinksotherthanm
and
j. Letp'^
—
K—lm{x^^)
for
some K.
Considerchangingthe pricesp^^ and
p^^
suchthatthenew
profitisf[g
=
Us
+
iK-
UxZ^
-
e))(x°^-
e)+
e{K
-
l,{e)).Note
thate units of flow aremoved
from linkm
to link j suchthat the flows ofother links
remain
thesame
atthenew
WE.
Hence,thechangeintheprofitisn,
-
n,
=
iUx^^)
-
Uxl^
-
e))x^^+
6(/„(x°^-
e)-
Z,(e))).Since /^(a;^^)
>
0, ecan be chosensufficientlysmallsuchthattheaboveisstrictlypositive, contradictingthefactthat {jp^,x'-'^) isan
OE.
Q.E.D.
The
followingexample
showsthatAssumption
2 cannotbe dispensed withfor part(b) ofthis
lemma.
Example
4
Consider athreehnk
network withtwo
providers,where
provider 1owns
links 1
and
3and
provider2owns
link2. Letthe total flowbe
d=
1and
the reservationutilitybei?
=
1.Assume
that the latency functions are givenby
^1(xi)== 0, ^2 (2^2)
=
2:2, h{x-i)=
ax-i,for
some
a>
0.Any
pricevector (pi,P2,P3)=
(2/3,1/3,6)with6>
2/3and
{xi,X2,X3)=
(2/3, 1/3,0) isa purestrategy
OE,
so^3X3=
contraryto part (b)ofthelemma. To
see
why
thisisanequilibrium, notethatprovider 2isclearlyplayingabest response.Moreover,in thisallocationIIi
=
4/9.We
canrepresentany
deviation of provider 1by
(pi,P3)
=
(2/3-5,2/3-ae-5),
for
two
scalarseand
5,whichwillinduceaWE
of(xj,X2,X3)=
(2/3+
6—
e,1/3—
5, e).
The
correspondingprofitofprovider1 atthisdeviationisHj=
4/9—
^^<
4/9,estab-lishingthat provider1is alsoplayinga bestresponse
and
we
haveapurestrategyOE.
We
next establish that, under an additional mild assumption, a purestrategyOE
willneverbeatapoint of non-differentiability of the latency functions.
Assumption
3There
existssome
sE
S
such that k is real-valuedand
continuouslydifferentiablefor alli
£
X, .Lemma
5 Let (p^-^,x°^)
be anOE
with min^{pf^
+
lj{x°^)]<
R
and pf
^xf
^
>
for
some
i. LetAssumptions
1,2and
3 hold.Then
where
lf{x^^)and
l~{xf^) arethe rightand
left derivatives ofthe functionU atxf^
respectively.
Since theproofof this
lemma
islong,itisgiveninAppendix
C.Note
thatAssumption
3cannot be dispensed within this
lemma.
Thisisillustratedinthenext example.Example
5 Consideratwo
Hnk
network. Letthe total flowhe d=
1and
thereserva-tion utilitybe/?
=
2.Assume
that thelatency functions aregivenby
^^(^^^^^(^)
=
\2(x-i)
otherwise.Itcan beverifiedthatthe vector (pf^,p^^)
=
(1, 1), with(xf^,x^^)
=
(1/2,1/2) isapurestrategy
OE, and
isat a point of non-differentiabihtyfor bothlatency functions.We
next provideanexplicitcharacterization oftheOE
prices,whichis essential inourefficiency analysisinSection 5.
The
proofisgiveninAppendix
D.Proposition
9 Let{p°^,x'^^) beanOE
suchthatpf^xf^
>
forsome
ie
1. LetAssumptions
1,2,and
3 hold.a)
Assume
thatmiuj{p^^
+
/j(x°^)}<
R. Then,for allsG
5
and
i€
2s,we
haver xf^/^(xf^), ifl'^{x°^)
=
forsome
j^
Z„
??""={
a:P^/'(xp^)
+ _^iM^4^,
otlierwise. (^6)b)
Assume
thatmiiij{p°^
+
lj{xf^)}=
R. Then, forallsG
5
and
i GXj,we
have
pr
>
^'^i^i^?'')- (17)Moreover,ifthere exists
some
iG
I
suchthat Z^=
{i} forsome
sG
5,thenpf^<xf^/,n:rf^)
+
^
^ 1 • (18)Ifthe latency functions li are all real-valued
and
continuously differentiable, thenanalysis of
Karush-Kuhn-
Tuckerconditionsfor ohgopolyproblem
[problem (82) inAp-pendixD] immediatelyyieldsthe followingresult:
Corollary
1 Let(p°^,x°^)
bean
OE
suchthatpf^xf^
>
forsome
iG
I. LetAs-sumptions1
and
2hold.Assume
alsothatkisreal-valuedand
continuouslydifferentiablefor alli. Then, forall s
G
5
and
i€ls, we
haveif/;(x°^)
=
Oforsomej ^J,
otherwise.
(19)
Thiscorollary alsoimpliesthatinthe
two
linkcasewithreal-valuedand
continuouslydifferentiablelatency functions
and
withminimum
effective cost lessthan R, theOE
pricesare
P?^-xf^{l[ixn
+
l2ixr))
(20)asclaimedinthe Introduction.
5
Efficiency
of
Oligopoly
Equilibria
This section contains our
main
results, providing tightbounds on
the inefficiency ofoligopolyequilibria.
We
take asourmeasure
ofefficiencythe ratio of thesocialsurplus oftheequilibriumflow allocation to thesocialsurplus of the social
optimum,
S{x*)/S{x^),where
x' refers tothemonopoly
or the oligopoly equilibrium [cf. Eq. (8)]. Section 3established that the flow allocation at a
monopoly
equilibrium is asocialoptimum.
Hence, in congestion
games
withmonopoly
pricing, there isno
efficiency loss.The
following
example
showsthatthisisnotnecessarilythe casewitholigopolypricing.Example
6 Consider atwo
linknetwork. Let thetotal flowbe d=
1and
thereservationutilityhe
R=
1.The
latency functions are givenby
3
/l(x)
=
0, l2{x)=
-X.The
uniquesocialoptimum
forthisexample
isx'^=
(1,0).The
uniqueME
(p^^,x'^^)is x'^'^
=
(1,0)and
p^^
=
(1,1).As
expected, the flow allocations at the socialoptimum
and
theME
arethesame.Next
consideraduopolywhere
eachofthese links isowned by
adifferent provider. UsingCorollary 1and
Lemma
3, itfollows thattheflow allocation atthe
OE,
x'-'^,satisfies/,(x?^)
+
x?^[/;(xf^)+
Ux^^)]
=
kix^)
+
x^Ux^)
+
Ux^'')]-Solvingthistogetherwith
xf ^
+
X2^
=
1 showsthat the flow allocation at theuniqueohgopoly equilibrium is x'~^^
—
(2/3,1/3).The
social surplus at the socialoptimum,
the
monopoly
equilibrium,and
the oligopoly equilibrium are givenby
1, 1,and
5/6, respectively.Before providinga
more
thoroughanalysis oftheefficiencyproperties oftheOE,
thenextpropositionprovesthat, asclaimedintheIntroduction
and
suggestedby
Example
6, achange in the
market
structure frommonopoly
toduopolyin atwo linknetworktypicallyreduces efficiency.
Proposition 10
Consider atwo
link network where eachhnk
isowned by
adiffer-ent provider. Let
Assumption
1 hold. Let {p^^,x'^^) be a pure strategyOE
suchthat
pp^xp^
>
forsome
i€
J
and
min^{pf'^
+
lj{xf^)}<
R. If Z'i(xf^)/xf^^
/^(x°^)/x^^,then S(xO^)/S(x^)
<
1 .Proof.
Combining
theOE
priceswiththeWE
conditions,we
havehix?")
+
x?^{l[ix?^)+
Ux2^))
^
hix^'^)+
xr{l[{xr)
+
Ux^)),
where
we
usethefactthatmiuj{p°^
+
l]{x'^^)}<
R. Moreover,we
canuse optimalityconditions (7)toprovethata vector (xf ,xf)
>
isasocialoptimum
ifand
onlyif/i(xf)
+
xf/'i(xf)=
hixl)+
xf/^(xf).Sinceri(xf^)/xf^
^
/^(x^^)/xp^,theresult follows.Q.E.D.
We
next quantify the efficiency of oligopoly equilibriaby
providing atightbound
on
theefficiency loss incongestiongames
witholigopolypricing.As we
haveshown
inSection4, such
games
do not always have a purestrategyOE.
Inthe following,we
firstprovide
bounds on
congestiongames
thathave purestrategyequilibria.We
next studyefficiencyproperties of
mixed
strategyequilibria.5.1
Pure
Strategy
Equilibria
We
consider pricecompetitiongames
thathave purestrategy equilibria(this setincludes,butissubstantially larger than,
games
withlinearlatency functions, see Section4).We
consider latency functions thatsatisfy
Assumptions
1,2,and
3. LetCi
denotethesetoflatency functionsfor which theassociated pricecompetition
game
has apurestrategyOE
and
the individual k's satisfyAssumptions
1,2,and
3.^^We
refertoan element ofthe set
Cj
by
{li}iei-Given
a parallellinknetwork with/Unks
and
latency functions{k}iei
S
-C/i IstOE{{li}) denote the setofflow allocations at anOE.
We
define theefficiencymetricat
some
x'^^G
OE{{li})asPV
x°^-T
l(x°^\x°^
where x^
isasocialoptimum
giventhelatency functions{li}i^jand
R
isthe reservationutility. In other words, our efficiency metric is the ratio ofthe social surplus in an
equihbriumrelative tothe siurplus in the social
optimum.
Following the literatureon
the "price ofanarchy",inparticular [25],
we
are interestedintheworstperformanceinan
oligopolyequihbrium,sowe
lookfora lowerbound on
inf inf^ r,{{U},x°^).
{'Oe£/
^oB^OEau))
We
firstprovetwo lemmas, which
reduce theset oflatency functions thatneedtobe considered in
bounding
the efficiency metric.The
nextlemma
allows us to use theoligopoly price characterization giveninProposition 9.
Lemma
6 Let {p'-'^,x^^)be
apurestrategyOE
suchthatpf^xf^
=
foralliG
Z.Then
x'~'^isa socialoptimum.
Proof.
We
firstshow
thath{xf^)
=
for alliel. Assume
that lj{x^^)>
forsome
;
€
1. Thisimphes
thatx^^
>
and
thereforepf^
=
0. Since lj{xf^)>
0, itfollowsby
Lemma
2thatforallx E
W[p),we
haveXj=
xf^. Considerincreasingpf^
tosome
small e
>
0.By
theupper
semicontinuityofW{p),
it follows that there existssome
e
>
sufficientlysmallsuchthatforallx6 W{e,p^^),
we
have\xj—
x^^\<
5forsome
5
>
0. Moreover,by
Proposition 2,we
have, for allx
€
W{e,p?.f), Xi>
xf^
foralli7^ j. Hence, theprofitofthe provider that
owns
linkjisstrictlyhigher at price vector{e,p^f) thanatp*^^,contradictingthefactthat
(p°^,x°^)
isanOE.
Clearly
xf^
>
forsome
jand
hencemin^gijpf^
+
k{x^'^)}=
pf^
+
lj{xf^)=
0,which
impliesby
Lemma
.3thatJ^tei^?^
=
d,. Using/j(xf^)=
0,and
G dU{xf^)
forall2,
we
haveR
-
kixf^)
-
xf
^5,.=
i?,V
iG
J,for
some
gi^€
dli{xf^). Hence,x°^
satisfiesthe sufficientoptimahty
conditions for asocial
optimum
[cf Eq. (7) withX^=
R],and
theresultfollows.Q.E.D.
The
nextlemma
allowsus toassume
without loss of generality thatRYliei^f
~
Y^iei^i(^i)^f
>
s-iidEiei^?^
=
dm.
thesubsequentanalysis.Lemma
7 Let {kjieiG
£/.Assume
that^^Moreexplicitly,Assumption2impliesthatifany
OE
(p'^'^,x'^^) associatedwith{li}iei hasxf^ >
andhixf^)
=
0,thenJ^=
{i}.either (i) Yliex^i(^f)^i
~
^^iei^f
^'^^some
socialoptimum
Xs, or (ii)E^eI^?^
<
^forsome
x°^
e0^{{h}).
Then
everyx°^
G
0E{{1,}) isasocialoptimum,
implyingthat rj{{li},x'^^)—
1.Proof.
Assume
that^i^xh{xf)xf
—
R^^i^jxf.
Sincex^
is asocialoptimum
and
everyx'-'^
G
OE{{k})
isafeasiblesolution to thesocialproblem
[problem(6)],we
have=
Y.^R -
k{xf))xf> J2iR -
h{x?''))xf'^,V x°^
€ O^iik}).
lei lex
By
the definition ofaWE,
we
havexf^
>
and
i?-
kixf^)
>
pf^
>
(wherepf^
istheprice oflinkiatthe
OE)
foralli. Thiscombined
with the precedingrelationshowsthat
x°^
is asocialoptimum.
Assume
nextthat Yl^^jxf^ <
dforsome
x'^^G
OE{{li}). Letp'-^^bethe associatedOE
price.Assume
thatp^^x'^^>
forsome
j€
I
(otherwisewe
aredone by
Lemma
6). Since
X^iei^?^
^
'^'^^
haveby
Lemma
3 thatminjgjjpj+
lj{x^^}=
R. Moreover,by
Lemma
4,itfollowsthatPjxf^>
foralliG
I. Hence,forallsG
5, ((pf^)iei,,a;°^)is
an
optimalsolution oftheproblemmaximize((p.).g^^,:c) y^^PiXj
subject to Pi
+
li{xi)=
R,V
iG
Xs,pf^
+
liix,)^R,
^lils.
Y.xO^<d.
Substitutingfor (p^ieis inthe above,
we
obtainmaximizex>o 2_^
(^
~
^i{^i))^iieis
subject to x^
&
Ti,M
i^I^,
^xf'Kd,
where
Tj=
{xj |pf^
+
li{xi)=
R}
iseitherasingleton or a closedinterval. Sincethis
isaconvex problem, using the optimality conditions,
we
obtainR
-
k{xf^)-
xf^gi,=0,
V
IG
X„ V
sG
5,where
gi^Gdli{xf^).By
Eq. (7), itfollowsthatx°^
is asocialoptimum.
Q.E.D.
This
lemma
implies that in findingalowerbound on
the efficiencymetric,we
canrestrict ourselves, without loss of generality, to latency functions {k}
G Cj
suchthatY^i&ik{xf)xf
<
RJ2iei^i
forsome
socialoptimum
x'^,and Xliei^?^
=
d for allx*^^
G
OE{{li}).By
thefollowinglemma,
we
canalsoassume
that YLiei^f~
'^•Lemma
8 Forasetoflatency functions{/,},gi,letAssumption
1hold. Let[p'~'^,x^^)be an
OE
and x^
bea socialoptimum.
Then
E-P'^E
Proof.
Assume
to arrive at acontradiction thatYliei^i -^ Yliei^i- Thisimpliesthat
x^^
>
Xj forsome
j.We
alsohave lj{x^^)>
lj{Xj). (Otherwise,we
would
havelj{xj)
—
l'j{Xj)=
0, whichyields acontradictionby
the optimalityconditions (7)and
thefactthat Yliei^f"^^)- Usingthe optimality conditions(2)
and
(7),we
obtainR
-
l,{xf^)-
pf
>R-
lj{x^)-
x^gi^,for
some
gi^€
dlj{Xj).Combining
the preceding with lj{xf^)>
lj{xj)and
p^^
>
xf'^lj{Xj'^) (cf.Proposition9),
we
seethatxf-l-ixf-Xx^g,,
contradicting
x^^
>
Xjand
completingthe proof.Q.E.D.
5.1.1
Two
Links
We
firstconsider a parallel linknetwork withtwo
linksowned by two
serviceproviders.The
nexttheorem
provides atightlowerbound
of5/6on
r2{{li},x'-'^) [cf. Eq. (21)].Startingwiththe two-linknetworkisuseful
two
reasons: first, thetwo-linknetworkavoidsthe additional layer ofoptimizationoverthe allocation of links to service providers
in characterizingthe
bound on
inefficiency;and
second,we
willprovethe resultforthegeneral case
by
reducingittotheproofofthe two-linkcase.Although
thedetails of theproofofthetheorem
are involved,thestructureisstraight-forward.
The
problem
offindingalowerbound
onr2{{k},x'-^^)isan
infinite-dimensionalproblem,sincetheminimizationisoverlatency functions.
The
prooffirstlower-boundsthe infinite-dimensional
problem by
theoptimalvalue ofafinite-dimensionaloptimiza-tion
problem
usingtherelationsbetween
the flows at socialoptimum
and
equihbrium,and
convexityofthe latency functions. Itthenshows
thatthe solutionwillinvolveoneofthe linkshaving zero latency. Finally, usingthisfact
and
the price characterizationfrom
Proposition 9, itreducestheproblem
of characterizing thebound on
inefficiencytoa simple minimization problem, withoptimalvalue 5/6.
An
intuitionfor thisvalueisprovidedbelow.
In the following,
we assume
withoutloss ofgenerality thatd—
1. Also recallthatlatency functionsin
C2
satisfyAssumptions
1,2,and
3.Theorem
1 Consider atwo
link networkwhere
each link isowned by
a differentprovider.
Then
r2{{k},x^^)
>
^,V
{kh=,,2e
C2,x^^
G O^iik}),
(22)and
thebound
istight,i.e.,thereexists {li}i=i,2€
^2and
x°^
G
OE{{li}) that attainsthelower
bound
inEq. (22).Proof.
The
prooffollowsanumber
ofsteps:Step 1:
We
are interestedinfindinga lowerbound
fortheproblem
inf inf r2({U,a:°^). (23)
Given {/j}
E
£2, letx°^
£ OE{{li})and
let x"^ be asocialoptimum.
By
Lemmas
7and
8,we
canassume
thatJ2i=i^?^
—
IZi=i^f—
^- ^^^^implies that there existssome
i such thatx^^
<
xf. Since theproblem
issymmetric,we
canrestrict ourselvesto {li]
€ £2
such that x'^^<
xf, i.e.,we
restrict ourselves to {li} G£2
such thatxf^
<
xf
—
eforsome
e>
0.We
claiminf inf r2({/.},x°^)>infrOf(6), (24)
where
we
defineproblem
(E^) asOF, ^
R
-
hy?'^-
hy?'^r2,t (e)
=
mmimize
,s, (,sy>o_
g g_
g g (Ei^-^fyf-^fyf
subject to If<yf{lfy,
i=
l,2, (25) li<y?^li, i=
l,2, (26) 'f+
yfaf)'=
^f+
yfOf)', (27) ^f+
yf(/f)'<i?, (28) 2h
+
i[{y?''-yl)<if,
(30)y2'"^>yf
+
e. (31)E:
Vi=
1, (32)+
{Ohgopoly
Equihbrium
Constraints}t, t=
1,2.Problem
(E')canbe viewedasafinitedimensionalproblem
thatcapturestheequilib-rium and
thesocialoptimum
characteristics oftheinfinitedimensionalproblem
given inEq. (23). Thisimpliesthat instead ofoptimizingoverthe entire functionk,
we
optimizeover the possible values ofli{-)
and
dli{-) atthe equilibriumand
the socialoptimum,
which
we
denoteby
li,l^,lf,(if)' [i.e., {if)' isavariablethat representsallpossibleval-ues ofgi^ Gdli{yf)].
The
constraints oftheproblem
guaranteethat these valuessatisfythenecessaryoptimality conditionsfora social
optimum
and
anOE.
In particular,con-ditions (25)