SLIDE 1
Pricing Public Goods for Private Sale Michal Feldman David Kempe - - PowerPoint PPT Presentation
Pricing Public Goods for Private Sale Michal Feldman David Kempe - - PowerPoint PPT Presentation
Pricing Public Goods for Private Sale Michal Feldman David Kempe (Harvard and Hebrew U) (U Southern California) Brendan Lucier Renato Paes Leme (Microsoft Research) (Microsoft Research) Some products in social networks typically benefit
SLIDE 2
SLIDE 3
SLIDE 4
SLIDE 5
SLIDE 6
Some products in social networks typically benefit not only the purchaser but also his friends (locally public goods).
SLIDE 7
Some products in social networks typically benefit not only the purchaser but also his friends (locally public goods).
- group video chat for Skype among collaborators
- poster printer among faculty in the department
- books among office mates
- snow-blower / gardening tools / … among neighbors
- shared infrastructure (public wi-fi) among companies
Yet, most of those products are typically sold privately. Our question: how to price goods over networks taking into account positive externalities ?
SLIDE 8
Related Work
- Public Goods: [Samuelson], [Bergstrom et al], …
- Networked Public Goods: [Bramoullé, Kranton],
[Bramoullé, Kranton, D’Amours], …
- Negative Externalities: [Jehiel, Moldovanu], [Jehiel et al],
[Brocas], …
- Positive Externalities: [Hartline et al], [Arthur et al],
[Akhalaghpour et al], [Anari et al], [Haghpanah et al], [Bhalgat et al]
- Pricing in Networks: [Candogan, Bimpikis, Ozdaglar], …
- Pricing Public Goods: [Bergstrom, Blume, Varian],
[Allouch], …
SLIDE 9
Model of Locally Public Goods
- agents embedded in a social network
- agent has value iid. Assume is atomless
- utilities: if is the set of allocated agents, then
where is the neighborhood of .
SLIDE 10
Model of Locally Public Goods
- agents embedded in a social network
- agent has value iid. Assume is atomless
- utilities: if is the set of allocated agents, then
where is the neighborhood of . Posted-prices game
- seller decides on prices
- each agent learns his value and the prices and
decides whether to buy or not
SLIDE 11
Posted-prices game
- seller decides on prices
- each agent learns his value and the prices and
decides whether to buy or not
SLIDE 12
Posted-prices game
- seller decides on prices
- each agent learns his value and the prices and
decides whether to buy or not
SLIDE 13
Posted-prices game
- seller decides on prices
- each agent learns his value and the prices and
decides whether to buy or not
SLIDE 14
Posted-prices game
- seller decides on prices
- each agent learns his value and the prices and
decides whether to buy or not
- equilibrium thresholds
- revenue
SLIDE 15
Posted-prices game Lemma: For any prices and distribution , there is a vector of equilibrium thresholds. If is uniform and the graph is regular, there is a symmetric equilibrium. Proof by fixed point arguments.
SLIDE 16
Posted-prices game Lemma: For any prices and distribution , there is a vector of equilibrium thresholds. If is uniform and the graph is regular, there is a symmetric equilibrium. Proof by fixed point arguments. Goal: Find prices to optimize/approximate the worst-case revenue over all equilibria, i.e.,
SLIDE 17
Posted-prices game Lemma: For any prices and distribution , there is a vector of equilibrium thresholds. If is uniform and the graph is regular, there is a symmetric equilibrium. Proof by fixed point arguments. Goal: Find prices to optimize/approximate the worst-case revenue over all equilibria, i.e., Questions: 1) What can we do if we have little or no knowledge of the network topology? 2) How does uniform (non-discriminatory) pricing perform ?
SLIDE 18
Three settings considered in our work 1) complete graph, regular distribution 2) d-regular graph, regular distribution 3) any graph, uniform distribution
SLIDE 19
Global Public Goods Theorem: For regular and , the uniform price guarantees in the worst equilibrium a constant fraction (1/8) of the revenue of any equilibrium at any price vector.
SLIDE 20
Global Public Goods Theorem: For regular and , the uniform price guarantees in the worst equilibrium a constant fraction (1/8) of the revenue of any equilibrium at any price vector. price vector asymmetric thresholds symmetric threshold
SLIDE 21
Global Public Goods Theorem: For regular and , the uniform price guarantees in the worst equilibrium a constant fraction (1/8) of the revenue of any equilibrium at any price vector. price vector asymmetric thresholds symmetric threshold
1
similar revenue (factor 2)
SLIDE 22
Global Public Goods Theorem: For regular and , the uniform price guarantees in the worst equilibrium a constant fraction (1/8) of the revenue of any equilibrium at any price vector. price vector asymmetric thresholds symmetric threshold
1
similar revenue (factor 2) Myerson (1 item, n players)
2
SLIDE 23
Global Public Goods Theorem: For regular and , the uniform price guarantees in the worst equilibrium a constant fraction (1/8) of the revenue of any equilibrium at any price vector. price vector asymmetric thresholds symmetric threshold
1
similar revenue (factor 2) Myerson (1 item, n players)
2
posted prices
3
SLIDE 24
Locally Public Goods (d-regular graphs) Theorem: For regular and d-regular graph , the uniform price , guarantees in the worst equilibrium a constant fraction of the revenue of worst equilibrium at any uniform price.
SLIDE 25
Locally Public Goods (d-regular graphs) Theorem: For regular and d-regular graph , the uniform price , guarantees in the worst equilibrium a constant fraction of the revenue of worst equilibrium at any uniform price. Show that this is necessary:
- unbounded gap between best best-case and
best worst-case revenue
- unbounded gap between discriminatory and non-
discriminatory pricing Proof uses the prophet price rather then the Myerson price.
SLIDE 26
Locally Public Goods (any graph) Theorem: For [0,1]-uniform and generic the uniform price guarantees a fraction of the revenue of worst equilibrium at any uniform price. Theorem: For uniform and generic , approximating the optimal revenue within a factor is NP-hard. I.e., we know a price that guarantees good revenue, yet knowing this value is hard.
SLIDE 27
Open Questions and Future Directions
- Imperfect substitutes
- Strength of social ties (weighted edges)
- Non-identical / Non-regular distributions
- Other objective functions
SLIDE 28
Open Questions and Future Directions
- Imperfect substitutes
- Strength of social ties (weighted edges)
- Non-identical / Non-regular distributions
- Other objective functions
Thanks !
SLIDE 29
SLIDE 30
Three settings:
1) complete graph, regular distribution 2) d-regular graph, regular distribution 3) general graph, uniform distribution
is regular iff is concave. Regular Distributions
SLIDE 31
Three settings:
1) complete graph, regular distribution 2) d-regular graph, regular distribution 3) general graph, uniform distribution
is regular iff is concave. Regular Distributions Myerson Price: s.t. Virtual value:
SLIDE 32
Global Public Goods Theorem: For regular and , the uniform price guarantees in the worst equilibrium a constant fraction (1/8) of the revenue of any equilibrium at any price vector. Proof: Inspired by a technique of [Chawla, Hartline, Kleinberg], we will compare the revenue with a posted price mechanism. Given we know that
SLIDE 33
Global Public Goods Now, consider as above and the corresponding symmetric equilibrium, Case 1.
SLIDE 34
Global Public Goods Now, consider as above and the corresponding symmetric equilibrium, Case 1.
SLIDE 35
Global Public Goods Now, consider as above and the corresponding symmetric equilibrium, Case 1.
SLIDE 36
Global Public Goods Now, consider as above and the corresponding symmetric equilibrium, Case 1.
SLIDE 37
Global Public Goods Now, consider as above and the corresponding symmetric equilibrium, Case 1.
SLIDE 38
Global Public Goods Now, consider as above and the corresponding symmetric equilibrium, Case 1.
SLIDE 39
Global Public Goods Now, consider as above and the corresponding symmetric equilibrium, Case 2.
SLIDE 40
Global Public Goods Now, consider as above and the corresponding symmetric equilibrium, Case 2.
SLIDE 41