Rubinstein Bargaining Model

The model's core interaction structure is a strictly alternating sequence of proposals. Agent 1 makes an initial offer on how to split a surplus (e.g., $1). Agent 2 can either accept, ending the game, or reject and immediately make a counteroffer. This turn-taking continues indefinitely or until acceptance. This formalizes real-world haggling and imposes a clear, recursive strategic structure.

A future payoff is worth less than a present one. Each agent has a discount factor (δ), where 0 < δ < 1. If an agent delays agreement by one period, any share they eventually receive is multiplied by δ. For example, with δ=0.9, $1 agreed next period is worth $0.90 today. This creates a cost to delay, forcing agents to balance holding out for a better offer against the erosion of value over time.

The model's solution is a Subgame Perfect Nash Equilibrium, where strategies are optimal at every possible point in the negotiation, including after any history of offers. This eliminates non-credible threats (e.g., "I'll never accept less than 80%"). The unique SPE provides a precise, predictable outcome based solely on the agents' discount factors and who moves first.

The equilibrium split directly reflects bargaining power derived from two sources:

• First-Mover Advantage: The agent who makes the initial offer typically secures a larger share.
• Relative Patience: An agent with a higher discount factor (more patient, δ closer to 1) is in a stronger position. The patient agent can afford to wait, pressuring the impatient agent to concede more. The solution formula quantifies this: if Agent 1 proposes first, their share = (1 - δ₂) / (1 - δ₁δ₂).

A key result is that agreement is reached immediately in the first period under perfect information and rationality. Despite the infinite horizon, no delay occurs in equilibrium because both agents can perfectly anticipate the future sequence of offers and the cost of delay. This contrasts with models of incomplete information, where delay is used as a costly signaling device.

In multi-agent systems, the Rubinstein model provides the theoretical bedrock for designing automated negotiation protocols. It informs:

• Agent reasoning: How a rational agent should evaluate offers based on its own and its opponent's discount rate.
• Protocol design: The efficiency of alternating-offers with time costs.
• Equilibrium analysis: A benchmark for evaluating the strategic properties of custom negotiation rules used by software agents.

A seminal cooperative game theory concept that provides an axiomatic solution to a two-player bargaining problem. It predicts the outcome of a negotiation where agents can achieve mutual gains from cooperation, assuming a threat point (disagreement outcome) and a set of feasible utility pairs. The solution maximizes the product of the players' utility gains relative to this threat point. Unlike the sequential Rubinstein model, the Nash solution is a single, static prediction based on fairness axioms like symmetry and independence of irrelevant alternatives.

A structured bilateral bargaining procedure where agents alternately make concessions from their previous offers. The protocol enforces that each new offer must be at least as favorable to the opponent as the previous one from that agent. Negotiation continues until an agreement is reached (offers intersect) or a deadline passes. This protocol provides a concrete, rule-based instantiation of the alternating-offers structure formalized in the Rubinstein model, often used in practical agent implementations where strict concession rules are required.

A mathematical representation of an agent's preferences, assigning a numerical value to each possible outcome or bundle of goods. In the Rubinstein Bargaining Model, each agent's utility for a share of the surplus is discounted over time, formalized as (U_i = \delta_i^t x_i), where (\delta) is the discount factor and (t) is time. The agent's strategic goal is to maximize this function. The discount factor is a critical parameter, representing patience; an agent with a higher (\delta) (closer to 1) has greater bargaining power.

The minimum price a seller is willing to accept or the maximum price a buyer is willing to pay, representing a private walk-away point. In strategic bargaining, this forms the agent's private valuation. While the Rubinstein model typically assumes a known, divisible surplus, introducing private reservation prices transforms the problem, often leading to inefficiencies or delays in agreement as agents engage in strategic screening. This connects the model to incomplete information bargaining scenarios.

A refinement of the Nash equilibrium concept central to the Rubinstein solution. It requires that the agents' strategies constitute a Nash equilibrium in every subgame of the original sequential game, eliminating non-credible threats. The Rubinstein model's unique SPE is derived through backward induction, where agents reason about the final period and work backwards. This equilibrium predicts immediate agreement, with the split determined by the agents' relative discount factors, showcasing the power of sequential rationality in protocol design.

The inverse of game theory, involving the design of negotiation protocols or 'games' so that the strategic interactions of self-interested agents lead to a socially desirable outcome. While the Rubinstein model analyzes a given alternating-offers protocol, mechanism design synthesizes protocols to achieve goals like efficiency, revenue maximization, or truth-telling. The revelation principle is a key tool here, stating that for any equilibrium of any complex mechanism (like alternating offers), an equivalent direct revelation mechanism exists.