Cognition and framing in sequential bargaining for gains and losses

Noncooperative game-theoretic models of sequential bargaining give an underpinning to cooperative solution concepts derived from axioms, and have proved useful in applications (see Osborne and Rubinstein 1990). But experimental studies of sequential bargaining with discounting have generally found systematic deviations between the offers people make and perfect equilibrium offers derived from backward induction (e.g., Ochs and Roth 1989). We have extended this experimental literature in two ways. First, we used a novel software system to record the information subjects looked at while they bargained. Measuring patterns of information search helped us draw inferences about how people think, testing as directly as possible whether people use backward induction to compute offers. Second, we compared bargaining over gains that shrink over time (because of discounting) to equivalent bargaining over losses that expand over time. In the games we studied, two players bargain by making a finite number of alternating offers. A unique subgame-perfect equilibrium can be computed by backward induction. The induction begins in the last period and works forward. Our experiments use a three-round game with a pie of $5.00 and a 50-percent discount factor (so the pie shrinks to $2.50 and $1.25 in the second and third rounds). In the perfect equilibrium the first player offers the second player $1.25 and keeps $3.75.