Bargaining in-Bundle over Multiple Issues in Finite- Horizon Alternating-Offers Protocol Francesco Di Giunta and Nicola Gatti Politecnico di Milano Milan,

Slides:



Advertisements
Similar presentations
N. Basilico, N. Gatti, F. Amigoni DEI, Politecnico di Milano Leader-Follower Strategies for Robotic Patrolling in Environments with Arbitrary Topology.
Advertisements

An Introduction to Game Theory Part V: Extensive Games with Perfect Information Bernhard Nebel.
Reaching Agreements II. 2 What utility does a deal give an agent? Given encounter  T 1,T 2  in task domain  T,{1,2},c  We define the utility of a.
Nash’s Theorem Theorem (Nash, 1951): Every finite game (finite number of players, finite number of pure strategies) has at least one mixed-strategy Nash.
This Segment: Computational game theory Lecture 1: Game representations, solution concepts and complexity Tuomas Sandholm Computer Science Department Carnegie.
3. Basic Topics in Game Theory. Strategic Behavior in Business and Econ Outline 3.1 What is a Game ? The elements of a Game The Rules of the.
Do software agents know what they talk about? Agents and Ontology dr. Patrick De Causmaecker, Nottingham, March
Consumer Sovereignty The interaction of supply and demand in the market mechanism.
Game Theoretical Insights in Strategic Patrolling: Model and Analysis Nicola Gatti – DEI, Politecnico di Milano, Piazza Leonardo.
Bargaining in Markets with One-Sided Competition: Model and Analysis Nicola Gatti DEI, Politecnico di Milano, Piazza Leonardo da.
Game Theory and Computer Networks: a useful combination? Christos Samaras, COMNET Group, DUTH.
M9302 Mathematical Models in Economics Instructor: Georgi Burlakov 2.5.Repeated Games Lecture
EC941 - Game Theory Lecture 7 Prof. Francesco Squintani
Negotiation A Lesson in Multiagent System Based on Jose Vidal’s book Fundamentals of Multiagent Systems Henry Hexmoor SIUC.
Preference Elicitation Partial-revelation VCG mechanism for Combinatorial Auctions and Eliciting Non-price Preferences in Combinatorial Auctions.
Short introduction to game theory 1. 2  Decision Theory = Probability theory + Utility Theory (deals with chance) (deals with outcomes)  Fundamental.
Game-theoretic analysis tools Necessary for building nonmanipulable automated negotiation systems.
A Introduction to Game Theory Xiuting Tao. Outline  1 st a brief introduction of Game theory  2 nd Strategic games  3 rd Extensive games.
ECO290E: Game Theory Lecture 4 Applications in Industrial Organization.
Dynamic Games of Complete Information.. Repeated games Best understood class of dynamic games Past play cannot influence feasible actions or payoff functions.
EC941 - Game Theory Prof. Francesco Squintani Lecture 8 1.
Sep. 5, 2013 Lirong Xia Introduction to Game Theory.
Planning under Uncertainty
UNIT II: The Basic Theory Zero-sum Games Nonzero-sum Games Nash Equilibrium: Properties and Problems Bargaining Games Review Midterm3/21 3/7.
Approximate and online multi-issue negotiation S.S. Fatima Loughborough University, UK M. Wooldridge N.R. Jennings University of.
Algorithmic Game Theory Nicola Gatti and Marcello Restelli {ngatti, DEI, Politecnico di Milano, Piazza Leonardo da Vinci 32, Milano,
Review: Game theory Dominant strategy Nash equilibrium
A Heuristic Bidding Strategy for Multiple Heterogeneous Auctions Patricia Anthony & Nicholas R. Jennings Dept. of Electronics and Computer Science University.
Agent Technology for e-Commerce Chapter 10: Mechanism Design Maria Fasli
Job Market Signaling (Spence model)
An introduction to game theory Today: The fundamentals of game theory, including Nash equilibrium.
An introduction to game theory Today: The fundamentals of game theory, including Nash equilibrium.
UNIT II: The Basic Theory Zero-sum Games Nonzero-sum Games Nash Equilibrium: Properties and Problems Bargaining Games Bargaining and Negotiation Review.
UNIT II: The Basic Theory Zero-sum Games Nonzero-sum Games Nash Equilibrium: Properties and Problems Bargaining Games Review Midterm3/23 3/2.
Towards Automated Bargaining in Electronic Markets: a Partially Two-Sided Competition Model N. Gatti, A. Lazaric, M. Restelli {ngatti, lazaric,
UNIT II: The Basic Theory Zero-sum Games Nonzero-sum Games Nash Equilibrium: Properties and Problems Bargaining Games Review Midterm3/19 3/5.
UNIT II: The Basic Theory Zero-sum Games Nonzero-sum Games Nash Equilibrium: Properties and Problems Bargaining Games Bargaining and Negotiation Review.
An introduction to game theory Today: The fundamentals of game theory, including Nash equilibrium.
Alternating-Offers Bargaining under One-Sided Uncertainty on Deadlines Francesco Di Giunta and Nicola Gatti Dipartimento di Elettronica e Informazione.
1 Bargaining & Markets u As before: Buyers and Sellers, δtp,δtp, δ t (1-p). u Matching: Seller meets a buyer with probability α. A buyer meets a seller.
MAKING COMPLEX DEClSlONS
Reading Osborne, Chapters 5, 6, 7.1., 7.2, 7.7 Learning outcomes
Sequences of Take-It-or-Leave-it Offers: Near-Optimal Auctions Without Full Valuation Revelation Tuomas Sandholm and Andrew Gilpin Carnegie Mellon University.
Evaluation of software engineering. Software engineering research : Research in SE aims to achieve two main goals: 1) To increase the knowledge about.
Dynamic Games of complete information: Backward Induction and Subgame perfection - Repeated Games -
Game Theory is Evolving MIT , Fall Our Topics in the Course  Classical Topics Choice under uncertainty Cooperative games  Values  2-player.
EC941 - Game Theory Prof. Francesco Squintani Lecture 5 1.
Bargaining Whoever offers to another a bargain of any kind, proposes to do this. Give me that which I want, and you shall have this which you want …; and.
Unions and Collective Bargaining Topic 4 Part I. Topic Outline Bargaining Models under Complete Information Applied to Collective Bargaining Nash’s Cooperative.
Game-theoretic analysis tools Tuomas Sandholm Professor Computer Science Department Carnegie Mellon University.
Designing the alternatives NRMLec16 Andrea Castelletti Politecnico di Milano Gange Delta.
Game Theory: introduction and applications to computer networks Game Theory: introduction and applications to computer networks Lecture 2: two-person non.
Bargaining Theory MIT Game Theory. Bargaining Theory Cooperative (Axiomatic) –Edgeworth –Nash Bargaining –Variations of Nash –Shapley Value Non-cooperative.
Topic 3 Games in Extensive Form 1. A. Perfect Information Games in Extensive Form. 1 RaiseFold Raise (0,0) (-1,1) Raise (1,-1) (-1,1)(2,-2) 2.
Price of Anarchy Georgios Piliouras. Games (i.e. Multi-Body Interactions) Interacting entities Pursuing their own goals Lack of centralized control Prediction?
An Extended Alternating-Offers Bargaining Protocol for Automated Negotiation in Multi-agent Systems P. Winoto, G. McCalla & J. Vassileva Department of.
Lecture 5 Introduction to Game theory. What is game theory? Game theory studies situations where players have strategic interactions; the payoff that.
Lecture 1 on Bargaining Setting the Agenda This lecture focuses on the well known problem of how to split the gains from trade or, more generally, mutual.
Chapter 6 Extensive Form Games With Perfect Information (Illustrations)
Definition and Complexity of Some Basic Metareasoning Problems Vincent Conitzer and Tuomas Sandholm Computer Science Department Carnegie Mellon University.
Bargaining games Econ 414. General bargaining games A common application of repeated games is to examine situations of two or more parties bargaining.
Quick summary One-dimensional vertical (quality) differentiation model is extended to two dimensions Use to analyze product and price competition Two.
Nash Bargaining Solution and Alternating Offer Games MIT Game Theory.
Lecture V: Bargaining Recommended Reading: Dixit & Skeath, Chapter 17 Osborne, Chapter 6.1, 16 Powell, In the Shadow of Power, Ch. 3.
Q 2.1 Nash Equilibrium Ben
Game Theory: The Competitive Dynamics of Strategy
Extensive-form games and how to solve them
Chapter 14 & 15 Repeated Games.
Chapter 14 & 15 Repeated Games.
Presentation transcript:

Bargaining in-Bundle over Multiple Issues in Finite- Horizon Alternating-Offers Protocol Francesco Di Giunta and Nicola Gatti Politecnico di Milano Milan, Italy

Summary Introduction to alternating-offers bargaining, open problems, and topic of the paper Review of the single-issue solution Basic ideas for our multi-issue solution Development of the multi-issue solution Conclusions and further work

Alternating-offers bargaining Two rational agents - a buyer b and a seller s – make offers and counteroffers in order to reach an agreement (e.g., on price, quality, quantity,… of a good to be sold) They have opposite interests and they both lose utility as time passes by Different settings: finite-horizon vs infinite-horizon single-issue vs multi-issue complete information vs incomplete information … The problem is: how should the two rational agents behave? Which should be their strategies?

Alternating-offers bargaining Game-theoretical analysis pioneered by [Stahl, 1972] and [Rubinstein, 1982] Long time interest in the game theory and in the artificial intelligence community The single issue problem with complete information is solved Slow further developments towards the solution of realistic models Main open problems: Incomplete information Multiple issues

Multi-issue problem Multi-issue bargaining protocols: Sequential: the issues are negotiated one by one In-bundle: all the issues are negotiated together Sequential bargaining does not assure Pareto- efficiency In-bundle bargaining is said to involve too much computations

Focus of our paper We focus on finite-horizon in-bundle alternating-offers bargaining with complete information We show that, for the most common kind of utility functions, the problem is indeed tractable We merge game-theoretical and linear/convex programming techniques

Review of the one-issue model The buyer b and the seller s act alternately at integer times Possible actions at time t are Make an offer (a real number, typically a price) Accept the opponent’s previous offer x: the outcome is (x,t) Exit the negotiation: the outcome is NoAgreement The utility function U b (U s ) of b (s) depends on her Reservation price RP b (RP s ) Deadline T b (T s ) Time discount factor δ b (δ s ) U b (x,t) = (RP b -x)(δ b ) t if t ≤ T b U b (x,t) = -1 if t > T b U s (x,t) = (x-RP s )(δ s ) t if t ≤ T s U s (x,t) = -1 if t > T U b (NoAgreement) = U s (NoAgreement) = 0

Review of the one-issue solution The appropriate notion of solution is subgame perfect Nash equilibrium The protocol is essentially a finite game, so the equilibrium can be found by backward induction: Call T = min {T b,T s } At time T the acting agent (say, s) would accept any offer with positive utility At time T-1 agent b would offer x * T-1 =RP s or accept any offer x such that U b (x,T-1) ≥ U b (x * T-1,T) At time T-2 agent s would offer x * T-2 such that U b (x * T-2,T-1) = U b (RP s,T) or accept any offer x such that U s (x,T-2) ≥ U s (x * T-2,T-1) … I.e., at each time point t, from T back, it is possible to recursively find the offer x * t that the acting rational agent would do if she would make an offer; such offers x * t (or possible irrational higher ones) are always accepted by the rational opponent. Therefore the agreement is achieved at the very beginning of the bargaining on the value x * 0

Towards the multi-issue solution The core of the single-issue solution is the calculation of the values x * t that one agent should offer at time t and the other should accept at time t+1 In the one-issue situation this is very easy Are there, in the multi-issue situation, tuples x * t of values that act somehow like these values x * t ? The answer, for a wide class of multi-issue utility functions, turns out to be yes Is the calculation of these values computationally tractable? Again, the answer is yes Is the attained agreement Pareto-efficient? Yes

Towards the multi-issue solution In single-issue bargaining, value x * t-1 is calculated from x * t as the value such that U i (x * t-1,t) = U i (x * t,t+1) where i is the agent that acts at time t I.e., x * t-1 is obtained as the one step “backward propagation” of x * t along the level curves of the utility function of agent i In multi-issue bargaining, instead, there is no unique “backward propagated” tuple x * t-1 = but an entire set of tuples X * t-1 which at time t are worth for agent i the same as x * t at time t+1

Basic idea for multi-issue bargaining We take as x * t-1 the tuple in X * t-1 that maximizes the utility of the agent acting at time t-1 For a wide range of utility functions, this can be done efficiently with linear/convex programming.

Multi-issue bargaining assumptions Linear multi-issue utility function of agent i: U i (x 1,…, x n,t) = ∑ j U j i (x j,t) if for each j U j i (x j,t) ≥ 0 U i (x 1,…, x n,t) = -1 otherwise where U j i (x j,t) = u j i (x j )(δ j b ) t if t ≤ T j i U j i (x j,t) = -1 otherwise where u j i are continuous, concave and strictly monotonic u j i are such that the agents have opposite preferences over each issue u j i are such that there are feasible agreements

Multi-issue bargaining solution T = min ji {T j i } is the global deadline of the bargaining Tuple x * T-1 = = where i is the agent that acts at time T To calculate x * t-1 from x * t (be s the agent that acts at time t) Calculate the set X * t-1 of tuples which at time t are worth the same as x * t at time t+1 for agent s Use linear/convex programming to calculate x * t-1 as the value in X * t-1 that maximizes the utility of agent b

Multi-issue bargaining solution Be σ* the following strategy profile: At time T accept any offer that has nonnegative value At time t<T accept any offer x such that agreement (x,t) has utility greater or equal to (x * t-1,t+1) and otherwise counteroffer x * t

Main results It can be shown that Strategy σ* is the unique subgame perfect equilibrium of the protocol The calculation of σ* is linear with T and polynomial with the number of issues With strategy profile σ*, the agreement is achieved immediately and is Pareto-efficient

Conclusions In this paper we have shown that complete information multi-issue bargaining is tractable, despite what is usually believed, for a wide (and the most common) range of utility functions and for the best known bargaining protocol Further work will deal with the incomplete information problem

Finally Thank you for your kind attention