Helpful context:

Two players want to split a dollar. They take turns making offers. If player 1’s offer is rejected, player 2 makes a counter-offer in the next period - but by then, the dollar has shrunk, because waiting is costly. Rubinstein (1982) proved that this infinite-horizon alternating-offers game has a unique subgame-perfect equilibrium, determined entirely by the players' discount factors. The solution is striking: rational players agree immediately, in the first period, and the entire shrinkage never actually occurs.

The result matters for two reasons. First, it gives a precise answer to a question that naive bargaining intuitions leave open: who gets more, and by how much? The answer depends only on patience - the more patient player extracts a larger share. Second, as the time between rounds shrinks to zero, the equilibrium payoffs converge to the Nash bargaining solution. This provides a strategic foundation for Nash’s axiomatic solution: it is not merely a normative ideal satisfying certain consistency axioms, but the outcome that fully rational, self-interested agents produce through an explicit procedure.

The mathematical engine is stationarity. The subgame that begins in any odd period (player 1 to offer) is structurally identical to the original game. This recursive structure collapses the infinite-horizon game into a system of two equations in two unknowns, which can be solved in closed form.

The Setup

Two players, $i \in \{1, 2\}$, bargain over a surplus normalized to 1. The game proceeds as follows. In period 1, player 1 proposes a split $(x, 1-x)$ with $x \in [0,1]$; player 2 accepts or rejects. If player 2 accepts, the game ends with payoffs $(x, 1-x)$. If player 2 rejects, the game moves to period 2, where player 2 proposes $(1-y, y)$; player 1 accepts or rejects. This alternation continues indefinitely.

Players discount future payoffs: player $i$ has discount factor $\delta_i \in (0,1)$, so a payoff of $v$ received in period $t$ is worth $\delta_i^{t-1} v$ today. Delay is costly to both players. A player who can credibly commit to waiting longer holds stronger bargaining power, because the other side loses more from delay.

There is no exogenous deadline. If no agreement is ever reached - if every offer is rejected forever - both players receive 0. This is the disagreement outcome.

Subgame Perfect Equilibrium

The solution concept is subgame perfect equilibrium (SPE): a strategy profile that constitutes a Nash equilibrium in every subgame, including those that are never reached on the equilibrium path. SPE rules out non-credible threats - a player cannot credibly threaten to reject an offer that is strictly better than what they could obtain by continuing, because in the subgame following that rejection, the threat would not be carried out.

The key structural observation is stationarity: the subgame beginning in any odd period, with player 1 to offer, is isomorphic to the original game. Similarly, the subgame beginning in any even period is isomorphic to the original game with the players' roles swapped. This means equilibrium strategies can be taken to be stationary - the same in every period of the same type.

Theorem (Rubinstein 1982). The alternating-offers game has a unique SPE. Player 1 proposes $(x^, 1 - x^)$ in period 1 and player 2 accepts immediately, where:

$$x^* = \frac{1 - \delta_2}{1 - \delta_1 \delta_2}, \qquad 1 - x^* = \frac{\delta_2(1 - \delta_1)}{1 - \delta_1 \delta_2}$$

Derivation. Let $x^$ denote player 1’s equilibrium share in any subgame where player 1 proposes, and let $y^$ denote player 1’s equilibrium share in any subgame where player 2 proposes (so player 2 receives $1 - y^*$).

When player 2 faces player 1’s offer, player 2 knows that if they reject, they will propose next period and receive $1 - y^$ in that period, worth $\delta_2(1 - y^)$ today. Player 2 therefore accepts any offer giving them at least $\delta_2(1 - y^)$. Player 1, maximizing their own share, offers player 2 exactly $\delta_2(1 - y^)$, keeping: $$x^* = 1 - \delta_2(1 - y^*) \qquad \text{(i)}$$

When player 1 faces player 2’s offer, player 1 knows that if they reject, they will propose next period and receive $x^$, worth $\delta_1 x^$ today. Player 2, minimizing what they concede, offers player 1 exactly $\delta_1 x^$: $$y^ = \delta_1 x^* \qquad \text{(ii)}$$

Substituting (ii) into (i): $x^* = 1 - \delta_2(1 - \delta_1 x^) = 1 - \delta_2 + \delta_1 \delta_2 x^$, giving $x^(1 - \delta_1 \delta_2) = 1 - \delta_2$, and therefore: $$x^ = \frac{1 - \delta_2}{1 - \delta_1 \delta_2}$$

Player 2’s share is $1 - x^* = \frac{\delta_2(1 - \delta_1)}{1 - \delta_1\delta_2}$, and player 1’s share when player 2 proposes is $y^* = \frac{\delta_1(1 - \delta_2)}{1 - \delta_1\delta_2}$.

Uniqueness follows from showing that these are the only values consistent with SPE - any deviation from these proposals would either be immediately rejected or leave payoffs on the table, contradicting equilibrium.

What the Solution Says

First-mover advantage. Player 1 receives $x^* = \frac{1-\delta_2}{1-\delta_1\delta_2}$ and player 2 receives $\frac{\delta_2(1-\delta_1)}{1-\delta_1\delta_2}$. Since $x^* > 1 - x^$ iff $1 - \delta_2 > \delta_2(1 - \delta_1)$ iff $1 - \delta_2 > \delta_2 - \delta_1\delta_2$ iff $1 > \delta_2(2 - \delta_1)$, player 1 has an advantage when players are impatient. For symmetric discount factors $\delta_1 = \delta_2 = \delta$: $x^ = \frac{1}{1+\delta} > \frac{1}{2}$ for all $\delta < 1$.

Patience and power. The comparative statics are clean: $$\frac{\partial x^}{\partial \delta_1} = \frac{\delta_2(1-\delta_2)}{(1-\delta_1\delta_2)^2} > 0, \qquad \frac{\partial x^}{\partial \delta_2} = -\frac{1 - \delta_1}{(1-\delta_1\delta_2)^2} < 0$$

Player 1’s share increases in their own patience and decreases in player 2’s patience. Patience is bargaining power.

Vanishing first-mover advantage. As $\delta_1, \delta_2 \to 1$, both shares approach $\frac{1}{2}$. Very patient players split equally; the first-mover advantage disappears.

Immediate agreement. On the equilibrium path, player 2 accepts player 1’s first-period offer. No delay ever occurs in equilibrium, even though the game allows infinite delay. This is because any agreement reached in a later period is worse for both parties (due to discounting) than the same agreement reached in period 1 - and rational players anticipate this.

Connection to Nash Bargaining

Represent discount factors as $\delta_i = e^{-r_i \Delta}$, where $r_i > 0$ is player $i$’s continuous-time discount rate and $\Delta > 0$ is the length of each period. As $\Delta \to 0$:

$$x^*(\Delta) = \frac{1 - e^{-r_2 \Delta}}{1 - e^{-(r_1+r_2)\Delta}} \xrightarrow{\Delta \to 0} \frac{r_2}{r_1 + r_2}$$

Player 1’s limiting share is $\frac{r_2}{r_1+r_2}$ and player 2’s is $\frac{r_1}{r_1+r_2}$. For symmetric discount rates $r_1 = r_2$, both get $\frac{1}{2}$.

This is exactly the Nash bargaining solution applied to the problem with disagreement point $d = (0,0)$ and feasible set $S = \{(s_1, s_2) : s_1 + s_2 \leq 1, s_i \geq 0\}$. The Nash solution maximizes $s_1 s_2$ over $S$, giving $s_1 = s_2 = \frac{1}{2}$ in the symmetric case. With asymmetric bargaining weights $\alpha_1 = \frac{r_2}{r_1+r_2}$, $\alpha_2 = \frac{r_1}{r_1+r_2}$, it maximizes $s_1^{\alpha_1} s_2^{\alpha_2}$, recovering the Rubinstein limit.

The significance: the Nash bargaining solution, derived axiomatically with no reference to a protocol, is the equilibrium outcome of explicit strategic bargaining as the frictions (time between rounds) vanish. The two approaches - axiomatic and strategic - converge to the same answer.

Summary

Concept Value
Player 1’s equilibrium share $\dfrac{1 - \delta_2}{1 - \delta_1\delta_2}$
Player 2’s equilibrium share $\dfrac{\delta_2(1 - \delta_1)}{1 - \delta_1\delta_2}$
Agreement timing Period 1 (immediate)
First-mover advantage Yes; $x^* = \frac{1}{1+\delta} > \frac{1}{2}$ for symmetric $\delta < 1$
Effect of patience More patient player captures larger share
Limiting solution ($\Delta \to 0$) Nash bargaining solution; equal split when $r_1 = r_2$
Uniqueness Unique SPE via stationarity and backward-induction argument

Read next: