2.2 Priing Internet aess in the presene of user loyalty
2.2.6 Experimental evaluation
We have developed a simulator to study the impat of the novel loyalty models. Simulations
analyze the behavior of ompeting loal ISPs. Eah ISP has some end users, a share of the
market. We suppose that the overall demand funtion for their servies is onstant: no users
enter or leave the market. This model is relevant to a saturated market, where everybodyan
aordtohaveInternetonnetivity,andInternetonnetivityisamust(seetheinelastiasein
Setion2.2.4fordetails). Wesupposethattheusermarketisinnitelydividableamongthexed
number of ISPs. The total end user market is normalized to 1,e.g., if there aretwo ISPs with
equal market share,thenboth ofthem have a marketshare of 0.5. ISPs ompetefor ustomers
bysettingtheir aesspriesineahround. Thepriingshemeusedis at-rateandwe assume
ahomogeneoususerpopulation(single,ommon reservationprie). ThelowestprieanISPan
set for a round is 0, while the highest prie is 100, whih orresponds to the reservation value
ommon to allusers. Note, that these priesaresymboli, representing thepossible lowestand
highestprie an ISPwould onsider.
Before the rst round, the loyalty model, the threshold
L
, the disount fatorθ
for futureprots and initial market shares are set. In eah round, end users may migrate from their
respetive provider with regard to the applied loyalty model (see Setion 2.2.5). Swithers
hoose their next ISP aording to two possible migration poliies: informed and greedy. The
ISP.For simpliity weassumethat ifthere aretwominimally-pried ISPs, halfofthemigrating
users joins one and the other half joins the other one. The greedy poliy represents a
partly-informeduserbase,whereusersswithtosome heaperISPtheyknowof. ThiswayalltheISPs
who announeaprie lower than their urrent ISPdraw someusers away.
Ineahround theISPs tryto maximizetheirinstant prot. Every ISPusesthesamesimple
andgreedystrategy. Theysupposethatthepriesoeredlastbytheirompetitionstaythesame
for the next round. Withthis in mind,they alulate their projeted marketshare hange and
prot by probing all possible priesthey an set, and nally, they hoose theprie thatwould
maximize their protinthe next round andthen playit.
The results presented in this setion are intended to point out some interesting issues
on-erning the impat of user loyalty on the priing ompetition of ISPs in dierent senarios. In
senarios where the stohasti loyalty model was used, every data point represents an average
of 10 simulation runs. Furthermore, parameter
σ
(standarddeviation) was set to0.5
; this wayvariations due to the stohasti behavior an be visualized eetively. The simulator ode is
availablefor download [45℄.
Referene senario
First, we lay down a referene senario whih serves as a baseline for theupoming
investiga-tions. Theparameter settings are: deterministi loyaltymodel, fully-informed migration poliy,
threshold
L = 1
(zero-level user loyalty and no administration onstraint for the ISPs), equal initial market sharesand disount fatorθ = 0.995
. Resultsfor two, three and tenISPs an beseenin Figure2.3.
Inthe aseof two ISPs,theplayers donot dare toinrease their pries, beause theywould
lose all of their ustomers if they did so. On the other hand, it is not worth dereasing the
pries, beause the prot inrease aused by gaining a larger portion of the user base does not
ompensate for the loss aused by dereasing the prie. Therefore, pries, prots and market
sharesstay thesameduring the game.
However,withthree ISPs,the situationhanges. Astwothirdsofthemarketbelongsto the
ompetition, itlooks promising to derease the prie,and grab alarger market share.
Unfortu-nately, allthe players think the same,whih leads toa prie war, andpriesfall quikly.
WithtenISPs,weseeasimilarproess. Whatseemssurprising, isthatthedereaseofpries
isslowerthan in the 3-player ase. Onewouldexpetthatlarger ompetitionleads to aquiker
prie war. However, withmore ISPs,the users attrated from eah ofthem add up,soyou an
grabthe same total numberof users witha smallerdierene inprie. On theother hand,you
want to make as muh out of your loyalustomers, as you an. These fatorslead to a slower
dereasein theprie.
0 20 40 60 80 100
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2
(a)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2
(b)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2
()
0 20 40 60 80 100
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(d)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(e)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(f)
0 20 40 60 80 100
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3 ISP4 ISP5 ISP6 ISP7 ISP8 ISP9 ISP10
(g)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3 ISP4 ISP5 ISP6 ISP7 ISP8 ISP9 ISP10
(h)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3 ISP4 ISP5 ISP6 ISP7 ISP8 ISP9 ISP10
(i)
Figure2.3: Referene senario: pries, protsandmarketsharesfor2,3and10ISPs
(determin-isti loyalty, fullyinformed,
L = 1
,equal initial marketshare)beause of at-rate priing; we hoose to depit both in the following senarios to failitate
understanding.
Degree of loyalty/swithing osts
Now, we set
L
below1
introduing user loyalty and administrative onstraints to the system.Comparing the results inFigure2.4 to thereferene senariothere aremajor dierenes.
First, looking at the 2-player ase, ISPs inrease their respetive pries to the maximum
in the rst round, and this prie stabilizes. This is beause users move muh slower now, so
whatever theprie dierene is, ISPs make more money by getting the most out of their loyal
users than by lowering their prie and attrating new ustomers up to the threshold. So both
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2
(a)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2
(b)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2
()
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(d)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(e)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(f)
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(g)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(h)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(i)
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3 ISP4 ISP5 ISP6 ISP7 ISP8 ISP9 ISP10
(j)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3 ISP4 ISP5 ISP6 ISP7 ISP8 ISP9 ISP10
(k)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3 ISP4 ISP5 ISP6 ISP7 ISP8 ISP9 ISP10
(l)
Figure 2.4: Pries, prots and market sharesfor 2 (
L = 0.05
), 3 (L = 0.05
andL = 0.15
) and10 ISPs (
L = 0.05
),deterministi loyaltyAlthoughtheproess startsthesamewaywiththreeISPs and asmallthreshold, itisworth
starting to dereaseyour pries fromthe maximumprie to grab a larger marketshare. When
all the players reah a relatively low prie, they see that they an make the highest prot by
taking advantage of their loyal userbase, and it is worth inreasing theprie to the maximum.
This proess is repeated periodially resulting in a sawtooth pattern. With a larger threshold
(meaninglessloyalusers),alargeruserbaseanbeattratedwithasinglederease. Thisenfores
larger dereases, and also brings the ruial prie lower. Note that the main harateristis of
theurvesremainthe same.
WithtenISPs(due tothesamereasonsaslistedinthereferenesenario)asmallerderease
inprieisenoughtoattratthesamenumberofusers,thereforethepriedereaserateisslower.
Stohastiloyalty
Letushange theloyaltymodeltothestohastitypewithstandarddeviationxedat
0.5
. Nowwesetparameter
L
todierentvaluesandobservethesimulation resultspresentedinFigure2.5.Notie how pries utuate wildly beause of the random-like number of swithers: the
sawtooth patternan still be reognized,but priesarenot drivento thehighestpossiblevalue.
This phenomenon has two reasons: rst, inany given round it may be more beneial for the
ISP to emphasize milking its loyal users than lowering its pries to attrat newustomers, and
vie versa; seond, every data point in Figure 2.5 is an average of ten simulation runs whih
smoothsout theurves,but is absolutelyneededto identify harateristi behavior.
Inthe ases, where
L
is low, pries, prots and market sharesutuate around theirdeter-ministi ounterparts (see Figure 2.4). However, when we assoiate zero administrative ost to
swithingISPs(
L = 1.0
) thingshange. There israndomnessinpriesandmarketsharesintherst few rounds, but priesfall to their minimumand staythere making market sharesfreeze
at a given level. This level is lose to the fair share, but ISP
1
has some disadvantage. Whathappenshereisthatthe stohastiityofuserloyaltyisnotsigniant enoughto movethepries
from their minimum, sothe marketonserves itsstate after theinitial utuations.
Inumbent vs. entrants
In an attempt to study an inumbent-newomer ISP setting we set theinitial market share of
ISP
1
to1.0
. We get bak to the deterministi loyalty model, and investigate how the loyalty level impats the prie war between the inumbent and the two entrants. Results are depitedinFigure 2.6. ISP
1
istheinumbent provider.It an be seen generally that entrant ISPs start grabbing the market from ISP
1
by settinglower pries. This makes the higher share ISP lower its pries aswell, resulting ina prie war.
0 20 40 60 80 100
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2
(a)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2
(b)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2
()
0 20 40 60 80 100
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(d)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(e)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(f)
0 20 40 60 80 100
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(g)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(h)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(i)
0 20 40 60 80 100
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3 ISP4 ISP5 ISP6 ISP7 ISP8 ISP9 ISP10
(j)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3 ISP4 ISP5 ISP6 ISP7 ISP8 ISP9 ISP10
(k)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3 ISP4 ISP5 ISP6 ISP7 ISP8 ISP9 ISP10
(l)
Figure 2.5: Pries, prots and marketsharesfor 2(
L = 0.15
),3 (L = 0.15
andL = 1.0
) and10ISPs (
L = 0.15
), stohastiloyalty0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(a)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(b)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
()
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(d)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(e)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(f)
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(g)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(h)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(i)
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(j)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(k)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(l)
Figure 2.6: Pries, protsand market sharesfor 3 ISPs (
L = 0.05, 0.20, 0.35, 1.0
), 1 inumbent,2 entrants, deterministiloyalty
dierent ompetition patterns. A highly loyal user population result in a low amplitude prie
utuationand fairlong-timemarketsharesfor allthree providers. A slightly moremobileuser
baseresults insharperpriehanges andperiodially evolvingmarketshares.
At
L = 0.35
,prieutuationsareattheextreme,alternatingbetweenveryhighandverylow pries. Thisis dueto thefatthatL
multipliedbythenumberof ISPsislarger than1
meaningthat the market an be ompletely redistributed in a single round. This results in a highly
unstable system. One would expet something similar in the ase when loyalty is non-existent
(
L = 1.0
). However, in this ase, a very quik prie war leads to very low and homogeneous pries. Note that beause the inumbent reahed the minimum prie last, its market sharebeame zero(refer to theformulaof prie sensitive loyalty inSetion 2.2.5).
Note that the stohasti loyalty model produes very similar ompetition in this senario
(resultsarenot shown).
Informed vs. uninformed users
In this senario we studied the dierene in market dynamis indued by fully and partially
informedusers. We usedthe stohastimodelwithstandarddeviation setto0.5,an
inumbent-newomersetting withthreeISPs. We ompare theresultsprodued withtheinformedand the
greedy migration modelinFigure2.7. ISP
1
istheinumbent provider.The lessons from these results arethe following. First, for a highly loyal market (rst and
third row), there is no big dierene in prie dynamis when omparing fully and partially
informed users. However, when users move easily (seond and fourth row), partially informed
users ausemore stablepriesand marketsharesonvergeswiftly toequilibrium. Thishappens
beausenotonlytheheapestISPgetsalltheusers,sothebenetofpriingbelowalltheothers
is lowerthan inthe fullyinformedase.
Prot in the long run
UserloyaltyalsoaetstheoverallprotofISPs. Here,weusethedeterministiloyaltymodelin
an inumbent-entrant setting andwedisount prots overtime witha disount fator of
0.995
.ISP
1
isthe inumbent provider. We an observe inFigure2.8 thatbothISPs do betteroverall,whenthereisahigh-levelofloyaltyamongusers. Furthermore,inFigure2.8(),asthetotallak
ofloyaltyandadministration onstraintsdrivesthe priestozeroinashorttime, both ISPsfail
to earnanyprotafter afew rounds.
In Figure 2.9, overall sum prots of two ISPs an be seen in time. Dierent lines denote
dierent levels of loyalty: the higher the threshold, the more users an swith providers at a
single step. Resultsshow that thehigher the level of loyalty (i.e., thelower thethreshold), the
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(a)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(b)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
()
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(d)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(e)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(f)
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(g)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(h)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(i)
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Prices
Round
ISP1 ISP2 ISP3
(j)
0 20 40 60 80 100
0 20 40 60 80 100 120
Profits
Round
ISP1 ISP2 ISP3
(k)
0 0.2 0.4 0.6 0.8 1
0 20 40 60 80 100 120
Market shares
Round
ISP1 ISP2 ISP3
(l)
Figure2.7: Pries,prots andmarketsharesfor 3ISPs withfullyinformed(
L = 0.05, 0.35
) andpartially informedusers (
L = 0.05, 0.35
), 1 inumbent, 2 entrants, deterministi loyalty0 500 1000 1500 2000 2500 3000 3500 4000 4500
0 20 40 60 80 100 120
Total profits
Round
ISP1 ISP2
(a)
0 500 1000 1500 2000 2500 3000 3500
0 20 40 60 80 100 120
Total profits
Round
ISP1 ISP2
(b)
0 20 40 60 80 100 120 140
0 20 40 60 80 100 120
Total profits
Round
ISP1 ISP2
()
Figure 2.8: Total individual prots for 2 ISPs (
L = 0.05, 0.30, 1.0
), 1 inumbent, 1 entrant,deterministi loyalty
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
0 20 40 60 80 100 120
Total profits
Round Threshold=0.1
Threshold=0.2 Threshold=0.3 Threshold=0.4
Figure2.9: Overall umulative prots of 2ISPs at dierent loyaltylevels
beingloyal to them,sine stronger loyalty results inhigher overallprots. Thisresultis inline
withreent empirialsurveys onuser loyalty [13℄.