Negotiation lab (3D)

Two players split a pie: patience, deadlines and outside options decide who gets what.

Switch to 2D

Presets

Playback

Players

Ultimatum game

Camera

Layers

Quality

Model check

The science

What you see

Two anonymous players, A (blue) and B (orange), split a pie of size 1 by alternating offers. Each round the proposer says how much to keep and the other accepts or rejects. After a rejection the pie shrinks: A's slices by the factor δA and B's by δB per round (impatience). Nodding means accepted, shaking the head means rejected. No real people, countries or conflicts are modelled.

Rubinstein alternating offers

With no deadline, the unique subgame-perfect equilibrium has agreement in round 1, with the first proposer keeping x₁ = (1 − δ₂)/(1 − δ₁δ₂). Backward induction on a finite horizon gives the same numbers as the deadline grows. The page checks the closed form against 200-round backward induction.

Patience is power

If a round lasts Δ and the impatience rates are ri (δi = e−riΔ), then as Δ → 0 the share tends to r₂/(r₁ + r₂): the Nash bargaining solution with bargaining power α = r₂/(r₁ + r₂) (Binmore, Rubinstein and Wolinsky).

Nash bargaining

With transferable utility and outside options b₁, b₂ the maximiser of (u₁ − b₁)α(u₂ − b₂)1−α on u₁ + u₂ = 1 is u₁ = b₁ + α(1 − b₁ − b₂). The page compares the numerical maximum with the formula.

Non-rational players

A stubborn player demands a fixed share and accepts nothing less; a conceding player lowers the demand each round down to 50 %. Two stubborn players with incompatible demands never agree and fall back on their outside options. Delay always destroys value because both sides discount.

Ultimatum game

One proposal, take it or leave it. A purely selfish responder would accept any positive offer, but many people dislike unequal splits. In the Fehr-Schmidt model the responder's utility is x − α·max(y − x, 0) − β·max(x − y, 0), so offers below α/(1 + 2α) are refused. If α varies between people (log-normal here) the proposer picks the offer that maximises (1 − x)·P(accept), which the page plots and checks by Monte Carlo.

Limits

Complete information, risk-neutral players, discounting as the only cost of delay, one-dimensional pie. Real negotiations involve uncertainty, reputation, commitment and many issues. Educational model, not advice about any real negotiation.

Loading model...
Loading Blender model...
Offers per round against the equilibrium
Equilibrium share against patience
Ultimatum game: acceptance and proposer payoff
Made with 🟠 Blender · Three.js