A Further Test of Noncooperative Bargaining Theory: Comment
通过两个实验检验Stahl/Rubinstein非合作讨价还价理论，发现该理论在超过两轮的博弈中被拒绝，而参与者更倾向于均分剩余。
A great deal of attention has recently been devoted to providing noncooperative game theory foundations to bargaining problems, (compare Ingolf Stahl, 1972, and Ariel Rubinstein, 1982). This approach has produced sharp predictions about the way surplus will be divided, and has given hope that a problem long thought to be insoluble might be resolved. This paper reports the results of two experiments designed to test the theory of bargaining developed by Stahl/Rubinstein.' Each experiment was designed to conform precisely to their rules of the game. We were particularly interested in how their solution, which is based on backward induction, would fare when bargaining was extended over more than two periods. We were also interested in how the outcomes would compare with the alternative hypothesis that bargainers tend to split a pie 50-50. Our work can also be seen as a response to the experiment reported by K. Binmore, A. Shaked, and J. Sutton (1985). They show that in a two-round game with alternating offers and a shrinking pie, experienced subjects chose offers consistent with the Stahl/Rubinstein theory. We duplicate their Game B result,2 but demonstrate that in games with more than two rounds the Stahl/Rubinstein theory is rejected. Thus to conclude on the basis of their results that subjects behave as gamesmen (i.e., in a manner consistent with the predictions of backward induction), would be premature. In experiments with varying numbers of rounds, our first players consistently demanded divisions which left their opponents with shares equal to the value of the second round pie.3 In a two-round game this behavior by definition yields demands consistent with the Stahl/Rubinstein theory. In games with more rounds it does not. The regularity of our subjects' behavior is most striking. In our experiments that went more than three rounds, the Stahl/Rubinstein division was strongly controverted. Furthermore, the divisions that were observed were remarkably similar across bargaining pairs and these divisions did not vary with experience or with the amount of money at stake.