Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Extending General Equilibrium Theory to the Digital Economy.

Slides:



Advertisements
Similar presentations
Chapter Price 6. Objectives: Students will learn… How the market establishes an equilibrium price How the equilibrium price balances supply & demand How.
Advertisements

6.853: Topics in Algorithmic Game Theory Fall 2011 Constantinos Daskalakis Lecture 16.
Chapter Twenty-Nine Exchange. u Two consumers, A and B. u Their endowments of goods 1 and 2 are u E.g. u The total quantities available and units of good.
4. The Problem of Exchange We consider now the development of competitive markets starting from 2-person barter exchange (direct exchange of goods) 4.1.
Combinatorial Algorithms for Market Equilibria Vijay V. Vazirani.
6.896: Topics in Algorithmic Game Theory Lecture 14 Constantinos Daskalakis.
Chapter Thirty Production. Exchange Economies (revisited) u No production, only endowments, so no description of how resources are converted to consumables.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Combinatorial Approximation Algorithms for Convex Programs?!
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Combinatorial Algorithms for Convex Programs (Capturing Market Equilibria.
1 Chapter 3 – Tools of Normative Analysis Public Finance McGraw-Hill/Irwin © 2005 The McGraw-Hill Companies, Inc., All Rights Reserved.
General Equilibrium Theory
Algorithmic Game Theory and Internet Computing
6.896: Topics in Algorithmic Game Theory Lecture 15 Constantinos Daskalakis.
6.853: Topics in Algorithmic Game Theory
CPS Topics in Computational Economics Instructor: Vincent Conitzer Assistant Professor of Computer Science Assistant Professor of Economics
Chapter Twenty-Nine Exchange. u Two consumers, A and B. u Their endowments of goods 1 and 2 are u E.g. u The total quantities available and units of good.
Microeconomic Theory Basic Principles and Extensions, 9e
Dynamic Spectrum Management: Optimization, game and equilibrium Tom Luo (Yinyu Ye) December 18, WINE 2008.
Algorithmic Game Theory and Internet Computing Amin Saberi Stanford University Computation of Competitive Equilibria.
Arbitrage in Combinatorial Exchanges Andrew Gilpin and Tuomas Sandholm Carnegie Mellon University Computer Science Department.
Types of Economic Systems Supply and Demand Price.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Algorithms for the Linear Case, and Beyond …
UNIT 3 – MARKETING Unit 3.03 Price and Distribute Products.
Supply and Demand Chapter 3 Copyright © 2010 by the McGraw-Hill Companies, Inc. All rights reserved. McGraw-Hill/Irwin.
Economics Basics Demand and Supply.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech New Market Models and Algorithms.
Chapter Thirty-Two Production. Exchange Economies (revisited)  No production, only endowments, so no description of how resources are converted to consumables.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Polynomial Time Algorithms For Market Equilibria.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Combinatorial Algorithms for Convex Programs (Capturing Market Equilibria.
6.896: Topics in Algorithmic Game Theory Lecture 13b Constantinos Daskalakis.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Polynomial Time Algorithms For Market Equilibria.
Approximating Market Equilibria Kamal Jain, Microsoft Research Mohammad Mahdian, MIT Amin Saberi, Georgia Tech.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Market Equilibrium: The Quest for the “Right” Model.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Joint work with Jugal Garg & Ruta Mehta Dichotomies in Equilibrium Computation: Market.
1 Exchange. 2 Two consumers, A and B. Their endowments of goods 1 and 2 are E.g. The total quantities available and units of good 1 units of good 2. and.
Comp 553: Algorithmic Game Theory Fall 2014 Yang Cai Lecture 23.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Primal-Dual Algorithms for Rational Convex Programs II: Dealing with Infeasibility.
2) Combinatorial Algorithms for Traditional Market Models Vijay V. Vazirani.
Chapter 5 Consumer surplus Household choice in input markets.
Demand: how much (quantity) of a product or service is desired by buyers Supply: How much of the good or service the market has to/can offer Law of Demand:
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani 3) New Market Models, Resource Allocation Markets.
© 2010 W. W. Norton & Company, Inc. 32 Production.
Chapter 32 Production. Exchange Economies (revisited) No production, only endowments, so no description of how resources are converted to consumables.
Algorithmic Game Theory and Internet Computing
Law of Demand ~ the amount of a product people will buy at different prices $20 $18 $16 $14 $12 $10 $8 $6 Demand Curve (D)
1 Production. 2 Exchange Economies (revisited) No production, only endowments, so no description of how resources are converted to consumables. General.
How Intractable is the ‘‘Invisible Hand’’: Polynomial Time Algorithms for Market Equilibria Vijay V. Vazirani Georgia Tech.
Ms. Kronlokken.  Microeconomics is a branch of economics that studies the behavior of individuals and firms in making decisions regarding the allocation.
© Thomson/South-Western ECONOMIC EDUCATION FOR CONSUMERS Slide 1 Consumer’s Role in the Economy Objectives: By the end of class, students will be able.
General Equilibrium (cont)
Chapter 32 Exchange.
Chapter Twenty-Nine Exchange.
Algorithmic Game Theory and Internet Computing
General Equilibrium (cont)
General Equilibrium (Social Efficiency)
Vijay V. Vazirani Georgia Tech
General Equilibrium (cont)
L13 General Equilibrium.
Lecture 2 Supply and demand
General Equilibrium (cont)
General Equilibrium (cont)
Chapter 33 Production.
Vocabulary Review Week 3.
Algorithmic Game Theory and Internet Computing
General Equilibrium (Social Efficiency)
General Equilibrium (Social Efficiency)
The Free Enterprise System
L12 General Equilibrium.
L12 General Equilibrium.
L13 General Equilibrium.
Presentation transcript:

Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Extending General Equilibrium Theory to the Digital Economy

Irving Fisher, 1891 Defined a fundamental market model Special case of Walras’ model

Several buyers with different utility functions and moneys.

W.l.o.g. assume 1 unit of each good

Several buyers with different utility functions and moneys. Equilibrium prices

Digital Goods Can be stored in the memory of a computer.

Pricing of Digital Goods From first principles! Music, movies, video games, … cell phone apps., …, web search results, …

Pricing of Digital Goods Music, movies, video games, … cell phone apps., …, web search results, … Once produced, supply is infinite!!

An interesting observation Any Pareto optimal allocation must give a copy of each digital good to each agent!

An interesting observation Any Pareto optimal allocation must give a copy of each digital good to each agent! Hence usual notion of equilibrium is not applicable!

Jain & V., 2010: Redefine “supply” and “demand” for digital goods and “map” digital economy onto AD-model. Proof of existence of equilibrium via Kakutani fixed point theorem. Efficient algorithm for one case.

Pricing of Digital Goods Music, movies, video games, … cell phone apps., …, web search results, …, even ideas! Once produced, supply is infinite!!

Idiosyncrasies of Digital Realm Staggering number of different goods, belonging to same genre, available with equal ease. E.g., iTunes has 11 million songs! App Store has 300,000 iPhone apps!

Game-Theoretic Assumptions Full rationality, infinite computing power

Game-Theoretic Assumptions Full rationality, infinite computing power e.g., song A for $1.23, song B for $1.56, …

Game-Theoretic Assumptions Full rationality, infinite computing power, and infinite patience! e.g., song A for $1.23, song B for $1.56, …

Game-Theoretic Assumptions Full rationality, infinite computing power: not meaningful! e.g., song A for $1.23, song B for $1.56, … Cannot price songs individually!

Uniform pricing has naturally emerged! Forerunners of digital goods, e.g., CD’s, DVD’s etc. iTunes  Started with $0.99 for each song.  Now has 3 categories: $0.69, $0.99 and $1.29

Model: categories of digital goods E.g., jazz music, children’s movies, classic movies. Uniform pricing of all goods in a category Generic good in a category: “song”

Categories of digital goods How to price categories? How to allocate goods from categories? Notion of equilibrium?

More idiosyncrasies … A person desiring 2 songs wants 2 different songs! 2 copies of same song no better than 1 copy!

More idiosyncrasies … A person desiring 2 songs wants 2 different songs! 2 copies of same song no better than 1 copy! Making a loaf of bread vs. making a copy of a song – very different!

Yet more … People have widely different liking for songs in a category & different people may have totally different likings!

Model: total orders Assume g categories of digital goods. : Set of songs in category j. These songs “define” category j. For each digital category j, each agent i has a total order over all songs in

Market Model Assume 1 conventional good: bread (1 unit), and g digital categories ( songs in category j). Each agent i has a coarse utility function over g digital categories and bread, and has money

Optimal bundle for i, given prices p Coarse allocation: Using and find optimal amounts: Detailed allocation: For each digital category, j, i gets her most favorite songs from, as given by

Copyright Law Only original owner/producer of a song can sell copies.

A basic difference In AD-model, any is a valid bundle. In our model: Suppose  {A, E} not a valid bundle  {A, B} is valid.

A basic difference In AD-model, any is a valid bundle. In our model: Suppose  {A, E} not a valid bundle  {A, B} is valid. Hence, our model cannot be “reduced” to AD-model.

The “mapping”

Each molecule of bread can be bought by at most one buyer. is replaced by Each song can be bought at most once by each buyer.

Supply and demand: conventional goods No deficiency: No surplus:

Supply and demand: digital goods No deficiency: No surplus:

Equilibrium (p, x) s.t. Each agent, i, gets optimal bundle (detailed allocation).  Does not demand more than songs in category j Market clears, i.e., all bread sold & at least full 1 copy of each song sold.

Equilibrium (p, x) s.t. Each agent, i, gets optimal bundle (detailed allocation).  Does not demand more than songs in category j Market clears, i.e., all bread sold & at least full 1 copy of each song sold. Cannot replace 1 by any other integer!

Market Model

Market Model Show: Equilibrium exists!

Theorem (Jain & V., 2010): Equilibrium exists. “market maker” consumes surplus songs, and produces songs to cover deficiency. At fixed point, MM does nothing!

Extensions of existence theorem Arrow-Debreu (exchange) model Introduce production in both models

Open More complex, realistic models, e.g., complementarity in production. Economics of ads in digital marketplaces. Model creation of new categories, extinction of old categories.

Algorithmic questions Efficient algorithms for remaining cases, especially first case with arbitrary no. of conventional goods + digital categories. Experimental verification & applications, e.g., pricing of new digital goods.

and more … Pricing of financial advice – need to sell to a few people at high price. Pricing of “innovative ideas” Pricing of drugs Other natural models that don’t need to rely on game theoretic assumptions of full rationality & infinite computing power.

Arrow-Debreu model with production Agent can use 8 hrs/day to make bread or write a song in any category (or any combination) total order over “defining” songs. Are owned by agents.

Arrow-Debreu model with production Feasible production of each agent is a convex, compact set in Agent i’s earning:  no. of units of bread produced  no. of copies of songs (owned or produced by i) sold Agent spends earnings on optimal bundle.

Equilibrium (p, x, y) s.t. Each agent, i, gets optimal bundle & “best” songs available in each category. Each agent, k, maximizes earnings, given p, x, y (-k)  Note: this aspect resembles Nash equilibrium! Market clears.

Market clears All bread produced is sold For each category j,  At least 1 full copy of each song in sold.  No agent demands more songs than those produced and those in

A question Q: What if a talented song writer wants to sell her songs at a premium? A: Negligible cost of copying. Hence important to capture market share! Our model automatically rewards best song writers.

Important clarification Our model is not postulating an entity like iTunes that prices categories. The market decides customary price for each category via supply & demand. Anyone ignoring market pricing norms is taking a risk of being ignored.