Bargaining and negotiation

Two parties can create value together only once they agree how to divide it. A government and a host community can turn a mineral deposit into revenue — or leave it in the ground behind a blockade. A union and an employer can keep a hospital open — or shut it with a strike. A trade bloc and a small exporter can unlock the gains from access — or trade on worse terms. Each is the same object: a surplus to split, and a disagreement point d fixing what each side gets if the deal collapses. Two frameworks predict the split — Nash’s axioms and Rubinstein’s alternating offers — and both say the same thing: your share depends less on what you demand than on what happens if you walk away.

The anatomy of a bargain

Formally a bargaining problem is a pair (F, d). F is the feasible set: every division the parties could jointly achieve, in each player’s utility (u₁, u₂). Its north-east boundary is the Pareto frontier, where nothing is wasted and one side gains only at the other’s expense. The disagreement point d = (d₁, d₂) is what each player receives if no deal is struck. A solution is a rule picking one point of F for each problem (F, d) — and the whole theory is about which point.

• The surplus (feasible set F): value that exists only if the parties agree — the rent from a producing mine, the gains from a trade deal, the output of a hospital that stays open.
• The threat point d: each side’s fallback if talks fail — a blockade, a strike, trading on WTO most-favoured-nation tariffs, a return to conflict. Disagreement is costly, so d sits well inside the frontier.
• The Pareto frontier: the efficient divisions of the surplus. Bargaining is a fight over where on it to land; the threat point decides who has the leverage to pull it their way.

The Nash Bargaining Solution

John Nash’s move in 1950 was to sidestep the haggling entirely. Instead of modelling who says what, he asked what properties any reasonable division should satisfy, wrote down four, and proved that exactly one rule satisfies all four at once. This is the axiomatic method: constrain the answer until a single candidate survives.

• Pareto efficiency: the solution lies on the frontier — no value is left on the table, since any waste could be redivided to make both better off.
• Symmetry: if the problem is symmetric — equal threat payoffs and a feasible set mirrored about the 45° line — the players receive equal payoffs.
• Invariance to affine transforms: rescaling a player’s utility by any positive affine transform (uᵢ → a·uᵢ + b, with a > 0) leaves the real outcome unchanged; utility has no natural scale or origin, so the split must not depend on them.
• Independence of irrelevant alternatives (IIA): if the chosen division survives a shrinking of F, deleting the other, unchosen options does not change it.

Nash’s theorem: exactly one solution satisfies all four axioms, and it is the division that maximises the product of the players’ gains over their threat point — the Nash product (u₁ − d₁)(u₂ − d₂), over all feasible (u₁, u₂). It is the product, not the sum: maximising a product punishes lopsided divisions, so the rule carries a strong fair-division flavour though no fairness axiom was assumed — symmetry and efficiency deliver it.

The rule is clearest with a transferable surplus, so the frontier is the line u₁ + u₂ = S. Substitute u₂ = S − u₁ into the Nash product and differentiate: the first-order condition (S − u₁ − d₂) − (u₁ − d₁) = 0 gives u₁* = (S + d₁ − d₂)/2, hence u₁* − d₁ = (S − d₁ − d₂)/2. Each side banks its own threat payoff dᵢ, then the two split the remaining surplus S − d₁ − d₂ straight down the middle.

Power-sharing fits the frame exactly. After Kenya’s disputed 2007 election the pie was executive authority and the disagreement point was continued violence — ruinous for both sides, which made the surplus from any agreement enormous. The 2008 National Accord, mediated by Kofi Annan’s African Union panel, divided that pie: the incumbent kept the presidency while a new prime minister’s office went to the opposition leader, with cabinet posts shared across the grand coalition. Where the line fell reflected each side’s threat point and outside options — the incumbent’s grip on the state against the challenger’s capacity to sustain the streets — and the mediator served as a commitment device that made the deal stick. The uglier the disagreement point, the larger the prize from not triggering it.

Worked example: dividing resource revenue

Take the sharpest African case. A producing mine or oilfield generates rents to be split between the national government (player 1) and the host county or community (player 2). If they agree, the operation runs and yields an annual surplus S = 100 illustrative units. If they do not, the community blockades the site — the threat point. A blockade destroys most of the value for both: the government salvages some through other revenue and partial operations (d₁), while the community, denied royalties and jobs, salvages almost nothing (d₂). Put numbers on it and turn the crank.

Players:  1 = national government,   2 = host county / community

Surplus from a deal:        S  = 100   (illustrative units of annual rent)
Threat point — a blockade destroys value for both:
  government salvages       d₁ = 20    (other revenue, partial operation)
  community salvages        d₂ = 0     (no royalties, no jobs, and it bears costs)

Net surplus to divide:      S − d₁ − d₂ = 100 − 20 − 0 = 80
Nash split — each side banks dᵢ, then takes half of the net surplus:
  u₁* = 20 + 80/2 = 20 + 40 = 60   → government
  u₂* =  0 + 80/2 =  0 + 40 = 40   → community

Division:  government 60,  community 40   (60 / 40)

The government takes 60, the community 40 — and the community’s 40 is entirely its share of the cooperative dividend, because its threat point is zero. A blockade hurts it as much as the government, so it enters the room with no floor to stand on. The split is not a judgement about desert; it reads who loses less when talks fail.

How a better outside option shifts the split

Now change the community’s fallback. A constitutional or statutory entitlement to a minimum share of resource revenue — the kind written into several African resource and revenue-sharing laws — means the community is owed something even in a dispute; or it lines up an alternative operator for the site. Either way its disagreement payoff rises from d₂ = 0 to, say, 30: the blockade is no longer costless to the state, because the community can now hold out with an income. Re-solve.

The community secures a credible fallback — a statutory royalty floor, or an
alternative buyer for the resource — so its disagreement payoff rises:

  d₂ : 0 → 30      (d₁ unchanged at 20)

Net surplus to divide:      S − d₁ − d₂ = 100 − 20 − 30 = 50
Nash split:
  u₁* = 20 + 50/2 = 20 + 25 = 45   → government
  u₂* = 30 + 50/2 = 30 + 25 = 55   → community

Division:  government 45,  community 55   (45 / 55)

Why the community gains +15 overall (40 → 55):
  higher threat point      +30   (it banks a better floor)
  smaller net surplus      −15   (net surplus 80 → 50, so each half falls 40 → 25)
  net change               +15

The division moves from 60/40 to 45/55 — a 15-unit swing — and the arithmetic says exactly why. The community banks 30 more at the floor, but as the net surplus shrinks from 80 to 50 its equal share falls by 15; the net effect is +15, and the government loses the mirror image. The lesson generalises: a better outside option shifts the split toward you one-for-one at the floor, net of your smaller half of a smaller dividend. But only a credible fallback counts — which is the next point.

Patience is power: the Rubinstein model

Nash tells you where a reasonable split lands, not how bargaining reaches it, and it takes d as given. Ariel Rubinstein’s 1982 model supplies the process. Two players divide a pie of size 1 by alternating offers: player 1 proposes; if player 2 rejects, a period passes and player 2 counter-offers; and so on. Delay is costly — each player discounts the future by a factor δᵢ ∈ (0, 1), so a share received one period later is worth only δᵢ times as much. δ is patience: near 1, waiting barely hurts; low, and you are desperate to settle now.

The game has a unique subgame-perfect equilibrium, and — remarkably — agreement is immediate, on the first offer. The logic is stationarity. Let x be player 1’s share when player 1 proposes. An offer is accepted only if it gives the responder at least the discounted value of rejecting and proposing next period. Because the two roles are structurally identical, player 1’s share when player 2 proposes is δ₁·x, so player 1 offers player 2 exactly δ₂(1 − δ₁·x) and keeps the rest: x = 1 − δ₂(1 − δ₁·x). Solving, x* = (1 − δ₂)/(1 − δ₁·δ₂), and player 2 gets δ₂(1 − δ₁)/(1 − δ₁·δ₂). No offer is ever rejected — the threat of delay does all the work while no delay occurs.

Rubinstein alternating offers, pie = 1.  First mover (player 1) receives:

        x* = (1 − δ₂) / (1 − δ₁ × δ₂)

(a) Equal patience,  δ₁ = δ₂ = 0.8
        x* = (1 − 0.8) / (1 − 0.8 × 0.8) = 0.20 / 0.36 ≈ 0.556
        → first mover 55.6%,  responder 44.4%   (a mild first-mover edge)

(b) Patient proposer, impatient responder:  δ₁ = 0.9,  δ₂ = 0.5
        x* = (1 − 0.5) / (1 − 0.9 × 0.5) = 0.50 / 0.55 ≈ 0.909
        → the patient first mover takes ≈ 91%

(c) Flip the patience:  δ₁ = 0.5,  δ₂ = 0.9
        x* = (1 − 0.9) / (1 − 0.5 × 0.9) = 0.10 / 0.55 ≈ 0.182
        → moving first no longer helps; the patient side takes ≈ 82%

(d) Both perfectly patient,  δ → 1
        x* → 1 / (1 + δ) → 1/2   (the first-mover advantage vanishes)

Two forces set the division. First, a first-mover advantage: proposing captures the round of discounting the responder would otherwise impose on you. Second, and stronger, relative patience: the higher-δ party can credibly say it can wait, and equilibrium hands it the larger share. Case (c) shows the ranking — an impatient party that moves first still loses badly to a patient responder. As the interval between offers shrinks toward zero, the first-mover advantage vanishes and the split converges on a Nash solution whose bargaining weights are the players’ relative patience. Nash’s equal split is the equal-patience case; Rubinstein shows what breaks it.

Outside options, deadlines, and commitment

Rubinstein’s threat point is the cost of delay, but players often also hold an outside option: a door to a fixed alternative that ends the bargaining. The outside-option principle says it changes the split only if it exceeds what the party would earn by staying — below that continuation value the threat to leave is empty and ignored; above it, it becomes a binding floor pulling the division toward its holder. Trade turns on this. A small economy negotiating access — under AGOA’s preferences or within AfCFTA — bargains over the gains from a deal against the status quo threat point: ordinary most-favoured-nation tariffs, or losing a preference it enjoys. An exporter with nowhere else to sell has a weak outside option and takes poor terms; one with diversified markets can credibly walk, and commands better ones.

This is why, paradoxically, tying your own hands can strengthen you — Thomas Schelling’s insight. A union with a strike fund, a government bound by a constitutional revenue-sharing floor, a delegation mandated not to concede below a line, a party that publicly burns its bridges: each removes its own flexibility and thereby makes its threat credible and its floor higher. Deadlines act the same way: a hard deadline — an expiring preference, a budget cliff, an election — concentrates the cost of delay and favours whoever suffers less from it. If it hurts you more, it weakens you; the skilled move is to arrange, credibly, that silence costs the other side more than it costs you.