Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Subscription Dynamics and Competition in
Communications Markets
Shaolei Ren, Jaeok Park, and Mihaela van der Schaar
Electrical Engineering DepartmentUniversity of California, Los Angeles
October 3, 2010
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 1 / 26
Introduction
Outline
1 Introduction
2 Model
3 User Subscription DynamicsEquilibrium AnalysisConvergence Analysis
4 Competition in Duopoly Markets
5 Illustrative Example
6 Conclusion
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 2 / 26
Introduction
Overview of Communications Markets
UsersService
Providers
QoS
Technology
Price
Revenue
Demand
Investment
Interaction among technology, users and service providers
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 3 / 26
Introduction
Our Work
UsersService
Providers
QoS
Technology
Price
Revenue
Demand
Investment
How does the technology influence the users’ demand and the service providers’
revenues?
• We consider a duopoly communications market.
• Given prices, how does QoS affect the subscription decisions (or demand) of users?
• How are prices determined through competition between the service providers?
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 4 / 26
Model
Outline
1 Introduction
2 Model
3 User Subscription DynamicsEquilibrium AnalysisConvergence Analysis
4 Competition in Duopoly Markets
5 Illustrative Example
6 Conclusion
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 5 / 26
Model
Model
……
NSP 1 NSP 2
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 6 / 26
Model
Model
……
NSP 1 NSP 2
Network model
• network service providers: S1 and S2
• continuum model: a large number of users
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 6 / 26
Model
Model
Service providers
• Si : price pi and fraction of subscribers λi (pi , p−i)
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 7 / 26
Model
Model
Service providers
• Si : price pi and fraction of subscribers λi (pi , p−i)
• utility (revenue): Ri (pi , p−i) = piλi (pi , p−i)
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 7 / 26
Model
Model
Service providers
• Si : price pi and fraction of subscribers λi (pi , p−i)
• utility (revenue): Ri (pi , p−i) = piλi (pi , p−i)
Users
• user k : uk = αkqi − pi if it subscribes to Si
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 7 / 26
Model
Model
Service providers
• Si : price pi and fraction of subscribers λi (pi , p−i)
• utility (revenue): Ri (pi , p−i) = piλi (pi , p−i)
Users
• user k : uk = αkqi − pi if it subscribes to Si
• αk follows a distribution with PDF f (α)
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 7 / 26
Model
Model
Service providers
• Si : price pi and fraction of subscribers λi (pi , p−i)
• utility (revenue): Ri (pi , p−i) = piλi (pi , p−i)
Users
• user k : uk = αkqi − pi if it subscribes to Si
• αk follows a distribution with PDF f (α)
assumptions on f (α)
• f (α) > 0 if α ∈ [0, β] and f (α) = 0 otherwise
• f (α) is continuous on [0, β]
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 7 / 26
Model
Model
Service providers
• Si : price pi and fraction of subscribers λi (pi , p−i)
• utility (revenue): Ri (pi , p−i) = piλi (pi , p−i)
Users
• user k : uk = αkqi − pi if it subscribes to Si
• αk follows a distribution with PDF f (α)
QoS model
• q1 is constant
• q2 = g(λ2), where g(λ2) ∈ (0, q1) is a differentiable and non-increasingfunction of λ2 ∈ [0, 1]
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 7 / 26
User Subscription Dynamics
Outline
1 Introduction
2 Model
3 User Subscription DynamicsEquilibrium AnalysisConvergence Analysis
4 Competition in Duopoly Markets
5 Illustrative Example
6 Conclusion
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 8 / 26
User Subscription Dynamics
User Subscription
• Discrete-time model {(λt1, λ
t2) | t = 0, 1, 2 · · · }
• Users’ belief model and subscription decisions
• naive (or static) expectation: every user expects that the QoS in the currentperiod is equal to that in the previous period (i.e., g̃k (λ
t2) = g(λt−1
2 ))• a user subscribes to whichever NSP provides a higher (non-negative) utility
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 9 / 26
User Subscription Dynamics
User Subscription
• Discrete-time model {(λt1, λ
t2) | t = 0, 1, 2 · · · }
• Users’ belief model and subscription decisions
• naive (or static) expectation: every user expects that the QoS in the currentperiod is equal to that in the previous period (i.e., g̃k (λ
t2) = g(λt−1
2 ))• a user subscribes to whichever NSP provides a higher (non-negative) utility
• Dynamics of user subscriptions
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 9 / 26
User Subscription Dynamics
User Subscription
• Discrete-time model {(λt1, λ
t2) | t = 0, 1, 2 · · · }
• Users’ belief model and subscription decisions
• naive (or static) expectation: every user expects that the QoS in the currentperiod is equal to that in the previous period (i.e., g̃k (λ
t2) = g(λt−1
2 ))• a user subscribes to whichever NSP provides a higher (non-negative) utility
• Dynamics of user subscriptions
if p1q1
> p2
g(λt−12 )
, then
λt1 = hd,1(λ
t−11 , λt−1
2 ) = 1− F
(
p1 − p2
q1 − g(λt−12 )
)
,
λt2 = hd,2(λ
t−11 , λt−1
2 ) = F
(
p1 − p2
q1 − g(λt−12 )
)
− F
(
p2
g(λt−12 )
)
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 9 / 26
User Subscription Dynamics
User Subscription
• Discrete-time model {(λt1, λ
t2) | t = 0, 1, 2 · · · }
• Users’ belief model and subscription decisions• naive (or static) expectation: every user expects that the QoS in the current
period is equal to that in the previous period (i.e., g̃k (λt2) = g(λt−1
2 ))• a user subscribes to whichever NSP provides a higher (non-negative) utility
• Dynamics of user subscriptions
if p1q1
≤ p2
g(λt−12 )
, then
λt1 = hd,1(λ
t−11 , λt−1
2 ) = 1− F
(
p1q1
)
,
λt2 = hd,2(λ
t−11 , λt−1
2 ) = 0.
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 9 / 26
User Subscription Dynamics Equilibrium Analysis
Equilibrium Analysis
• Stabilized fraction of subscribers will stabilize in the long run
Definition
(λ∗
1 , λ∗
2) is an equilibrium point of the user subscription dynamics in the duopolymarket if it satisfies hd,1(λ
∗
1 , λ∗
2) = λ∗
1 and hd,2(λ∗
1 , λ∗
2) = λ∗
2 .
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 10 / 26
User Subscription Dynamics Equilibrium Analysis
Equilibrium Analysis
• Stabilized fraction of subscribers will stabilize in the long run
Proposition (uniqueness and existence of (λ∗
1 , λ∗
2))
For any non-negative price pair (p1, p2), there exists a unique equilibrium point(λ∗
1 , λ∗
2) of the user subscription dynamics in the duopoly market. Moreover,(λ∗
1 , λ∗
2) satisfies
λ∗
1 = 1− F
(
p1
q1
)
, λ∗
2 = 0, ifp1
q1≤
p2
g(0),
λ∗
1 = 1− F (θ∗1 ) , λ∗
2 = F (θ∗1 )− F (θ∗2 ) , ifp1
q1>
p2
g(0),
where θ∗1 = (p1 − p2)/(q1 − g(λ∗
2)) and θ∗2 = p2/g(λ∗
2).
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 10 / 26
User Subscription Dynamics Equilibrium Analysis
Equilibrium Market Shares
0
1
2
3
0
1
2
30
0.2
0.4
0.6
0.8
1
p1
p2
λ*
λ1*
λ2*
• q1 = 2.5, g(λ2) = 1.2e−0.5λ2 , and α is uniformly distributed on [0, 1], i.e., fa(α) = 1 for α ∈ [0, 1].
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 11 / 26
User Subscription Dynamics Convergence Analysis
Convergence of User Subscription Dynamics
• Convergence is not always guaranteed
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 12 / 26
User Subscription Dynamics Convergence Analysis
Convergence of User Subscription Dynamics
• Convergence is not always guaranteed
Example: when the QoS of NSP S2 degrades fast w.r.t. the fraction of subscribers
1 suppose that only a small fraction of users subscribe to NSP S2 at period t andeach subscriber obtains a high QoS
2 a large fraction of users subscribe at period t + 1, which will result in a low QoS atperiod t + 1
3 a small fraction of subscribers at period t + 2
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 12 / 26
User Subscription Dynamics Convergence Analysis
Convergence of User Subscription Dynamics
• Convergence is not always guaranteed
Theorem
For any non-negative price pair (p1, p2), the user subscription dynamics convergesto the unique equilibrium point starting from any initial point (λ0
1, λ02) ∈ Λ if
maxλ2∈[0,1]
{
−g ′(λ2)
g(λ2)·
q1q1 − g(λ2)
}
<1
K,
where K = maxα∈[0,β] f (α)α. proof
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 12 / 26
User Subscription Dynamics Convergence Analysis
Illustration of Oscillation & Convergence
2 4 6 8 10 12 140
0.2
0.4
0.6
0.8
t
λ 2
λ2: g(λ
2)=e−2λ
2
2 4 6 8 10 12 140.3
0.32
0.34
0.36
0.38
0.4
t
λ 2
λ2: g(λ
2)=e−0.8λ
2
t = 2
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 13 / 26
User Subscription Dynamics Convergence Analysis
Illustration of Oscillation & Convergence
2 4 6 8 10 12 140
0.2
0.4
0.6
0.8
t
λ 2
2 4 6 8 10 12 140.3
0.32
0.34
0.36
0.38
0.4
t
λ 2
λ2: g(λ
2)=e−2λ
2
λ2: g(λ
2)=e−0.8λ
2
t = 3
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 13 / 26
User Subscription Dynamics Convergence Analysis
Illustration of Oscillation & Convergence
2 4 6 8 10 12 140
0.2
0.4
0.6
0.8
t
λ 2
2 4 6 8 10 12 140.3
0.32
0.34
0.36
0.38
0.4
t
λ 2
λ2: g(λ
2)=e−2λ
2
λ2: g(λ
2)=e−0.8λ
2
t = 4
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 13 / 26
User Subscription Dynamics Convergence Analysis
Illustration of Oscillation & Convergence
2 4 6 8 10 12 140
0.2
0.4
0.6
0.8
t
λ 2
2 4 6 8 10 12 140.3
0.32
0.34
0.36
0.38
0.4
t
λ 2
λ2: g(λ
2)=e−2λ
2
λ2: g(λ
2)=e−0.8λ
2
t = 5
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 13 / 26
User Subscription Dynamics Convergence Analysis
Illustration of Oscillation & Convergence
2 4 6 8 10 12 140
0.2
0.4
0.6
0.8
t
λ 2
2 4 6 8 10 12 140.3
0.32
0.34
0.36
0.38
0.4
t
λ 2
λ2: g(λ
2)=e−2λ
2
λ2: g(λ
2)=e−0.8λ
2
t = 6
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 13 / 26
User Subscription Dynamics Convergence Analysis
Illustration of Oscillation & Convergence
2 4 6 8 10 12 140
0.2
0.4
0.6
0.8
t
λ 2
λ2: g(λ
2)=e−2λ
2
2 4 6 8 10 12 140.3
0.32
0.34
0.36
0.38
0.4
t
λ 2
λ2: g(λ
2)=e−0.8λ
2
t = 7
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 13 / 26
User Subscription Dynamics Convergence Analysis
Illustration of Oscillation & Convergence
2 4 6 8 10 12 140
0.2
0.4
0.6
0.8
t
λ 2
2 4 6 8 10 12 140.3
0.32
0.34
0.36
0.38
0.4
t
λ 2
λ2: g(λ
2)=e−2λ
2
λ2: g(λ
2)=e−0.8λ
2
t = 15
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 13 / 26
Competition in Duopoly Markets
Outline
1 Introduction
2 Model
3 User Subscription DynamicsEquilibrium AnalysisConvergence Analysis
4 Competition in Duopoly Markets
5 Illustrative Example
6 Conclusion
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 14 / 26
Competition in Duopoly Markets
Cournot Competition
• We model competition between the NSPs using Cournot competition.
• each NSP chooses the fraction of subscribers independently• prices are determined such that the equilibrium market shares equate the
chosen quantities
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 15 / 26
Competition in Duopoly Markets
Cournot Competition
• We model competition between the NSPs using Cournot competition.
• each NSP chooses the fraction of subscribers independently• prices are determined such that the equilibrium market shares equate the
chosen quantities
• GC = {Si ,Ri(λ1, λ2), λi ∈ [0, 1) | i = 1, 2}
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 15 / 26
Competition in Duopoly Markets
Cournot Competition
• We model competition between the NSPs using Cournot competition.
• each NSP chooses the fraction of subscribers independently• prices are determined such that the equilibrium market shares equate the
chosen quantities
• GC = {Si ,Ri(λ1, λ2), λi ∈ [0, 1) | i = 1, 2}
• (λ∗∗
1 , λ∗∗
2 ) is a (pure) NE of GC (or a Cournot equilibrium) if it satisfies
Ri (λ∗∗
i , λ∗∗
−i ) ≥ Ri (λi , λ∗∗
−i ), ∀ λi ∈ [0, 1), ∀ i = 1, 2 .
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 15 / 26
Competition in Duopoly Markets
Existence of NE
Lemma
Suppose that f (·) is non-increasing on [0, β]. Let λ̃i (λ−i ) be a market share thatmaximizes the revenue of NSP Si provided that NSP S−i chooses λ−i ∈ [0, 1),i.e., λ̃i (λ−i ) ∈ argmaxλi∈[0,1) Ri(λi , λ−i ). Then λ̃i (λ−i ) ∈ (0, 1/2] for all
λ−i ∈ [0, 1), for all i = 1, 2. Moreover, λ̃i (λ−i ) 6= 1/2 if λ−i > 0, for i = 1, 2.
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 16 / 26
Competition in Duopoly Markets
Existence of NE
Lemma
Suppose that f (·) is non-increasing on [0, β]. Let λ̃i (λ−i ) be a market share thatmaximizes the revenue of NSP Si provided that NSP S−i chooses λ−i ∈ [0, 1),i.e., λ̃i (λ−i ) ∈ argmaxλi∈[0,1) Ri(λi , λ−i ). Then λ̃i (λ−i ) ∈ (0, 1/2] for all
λ−i ∈ [0, 1), for all i = 1, 2. Moreover, λ̃i (λ−i ) 6= 1/2 if λ−i > 0, for i = 1, 2.
• Implication
• when the strategy space is specified as [0, 1) and f (·) satisfies thenon-increasing property, strategies λi ∈ {0} ∪ (1/2, 1) is strictly dominated fori = 1, 2
• if a NE (λ∗∗
1 , λ∗∗
2 ) of G̃C exists, then it must satisfy (λ∗∗
1 , λ∗∗
2 ) ∈ (0, 1/2)2
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 16 / 26
Competition in Duopoly Markets
Existence of NE
Theorem
Suppose that f (·) is non-increasing and continuously differentiable on [0, β]. Iff (·) and g(·) satisfy some conditions (Eqn. 18 and Eqn. 19 in the paper), thenthe game G̃C has at least one NE.
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 17 / 26
Competition in Duopoly Markets
Existence of NE
Corollary
Suppose that the users’ valuation of QoS is uniformly distributed, i.e., f (α) = 1/βfor α ∈ [0, β]. If g(λ2) + λ2g
′(λ2) ≥ 0 for all λ2 ∈ [0, 1/2], then the game GC hasat least one NE.
• Interpretation
• if the elasticity of the QoS provided by NSP S2 with respect to the fraction ofits subscribers is no larger than 1 (i.e., −[g ′(λ2)λ2/g(λ2)] ≤ 1), the Cournotcompetition game with the strategy space [0, 1) has at least one NE
• the condition is analogous to the sufficient conditions for convergence in thatit requires that the QoS provided by NSP S2 cannot degrade too fast withrespect to the fraction of subscribers.
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 18 / 26
Illustrative Example
Outline
1 Introduction
2 Model
3 User Subscription DynamicsEquilibrium AnalysisConvergence Analysis
4 Competition in Duopoly Markets
5 Illustrative Example
6 Conclusion
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 19 / 26
Illustrative Example
Numerical Results
0 5 10 15 200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
t
(λ1,λ
2)
λ1
λ2
Figure: Dynamics of market shares under the best-response dynamics. Solid:g(λ2) = 1− λ2
8; dashed: g(λ2) = 1− λ2
2.
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 20 / 26
Illustrative Example
Numerical Results
0 5 10 15 200
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
t
(R1,R
2)
R1
R2
Figure: Iteration of revenues under the best-response dynamics. Solid: g(λ2) = 1− λ28;
dashed: g(λ2) = 1− λ22.
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 21 / 26
Conclusion
Outline
1 Introduction
2 Model
3 User Subscription DynamicsEquilibrium AnalysisConvergence Analysis
4 Competition in Duopoly Markets
5 Illustrative Example
6 Conclusion
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 22 / 26
Conclusion
Conclusion
Study the impacts of technologies on the user subscription dynamics
• constructed the dynamics of user subscription based on myopic updates
• showed that the existence of a unique equilibrium point of the usersubscription dynamics
• provided a sufficient condition for the convergence of the user subscriptiondynamics: the QoS provided by NSP S2 should not degrade too fast as moreusers subscribe
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 23 / 26
Conclusion
Conclusion
Study the impacts of technologies on the user subscription dynamics
• constructed the dynamics of user subscription based on myopic updates
• showed that the existence of a unique equilibrium point of the usersubscription dynamics
• provided a sufficient condition for the convergence of the user subscriptiondynamics: the QoS provided by NSP S2 should not degrade too fast as moreusers subscribe
Study the impacts of technologies on competition between the NSPs
• modeled the NSPs as strategic players in a non-cooperative Cournot game
• provided a sufficient condition that ensures the existence of at least one NEof the game
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 23 / 26
Conclusion
Selected References
• J. Musacchio and D. Kim, “Network platform competition in a two-sided market: implications to the netneutrality issue,” TPRC: Conference on Communication, Information, and Internet Policy, Sep. 2009.
• D. Bertsimas and G. Perakis, “Dynamic pricing: a learning approach,” Models for CongestionCharging/Network Pricing, 2005.
• G. Bitran and R. Galdentey, “An overview of pricing models for revenue management,” Manufacturing& Service Operational Management, vol. 5, no. 3, pp. 203-229, Summer 2003.
• F. Kelly, “Charging and rate control for elastic traffic,” European Transactions on Telecommunications,vol. 8, pp. 33-37, 1997.
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 24 / 26
Conclusion
Related Publications
• S. Ren, J. Park, and M. van der Schaar, “Dynamics of Service Provider Selection in CommunicationMarkets,” accepted and to appear in Proc. IEEE Globecom 2010.
• S. Ren, J. Park, and M. van der Schaar, “Subscription Dynamics and Competition in CommunicationMarkets,” in Proc. ACM NetEcon 2010.
• S. Ren, J. Park, and M. van der Schaar, “User Subscription Dynamics and Revenue Maximization inCommunication Markets,” UCLA Tech. Report, Aug. 2010 (available athttp://arxiv.org/abs/1008.5367).
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 25 / 26
Conclusion
Convergence of User Subscription Dynamics
Proof.
1 Show that
‖hd(λ1,a, λ2,a)− hd(λ1,b, λ2,b)‖∞
= K
[
−g ′(λ2,c)
g(λ2,c)·
q1q1 − g(λ2,c)
]
|λ2,a − λ2,b|
≤ κd ‖λa − λb‖∞ .
where κd = K ·maxλ2∈[0,1] {[−g ′(λ2)/g(λ2)] · [q1/(q1 − g(λ2))]}
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 26 / 26
Conclusion
Convergence of User Subscription Dynamics
Proof.
1 Show that
‖hd(λ1,a, λ2,a)− hd(λ1,b, λ2,b)‖∞
= K
[
−g ′(λ2,c)
g(λ2,c)·
q1q1 − g(λ2,c)
]
|λ2,a − λ2,b|
≤ κd ‖λa − λb‖∞ .
where κd = K ·maxλ2∈[0,1] {[−g ′(λ2)/g(λ2)] · [q1/(q1 − g(λ2))]}
2 If maxλ2∈[0,1]
{
− g ′(λ2)g(λ2)
· q1q1−g(λ2)
}
< 1K
, then the mapping is contraction!
back
Ren, Park, van der Schaar (UCLA) Multimedia Communications & Systems Lab October 2010 26 / 26