A test of the characteristic function and the Harsanyi function in N- person normal form sidepayment games (Q1093579)

This paper reports the results of a laboratory experiment investigating sidepayment games represented in normal form. Attempts to predict payoff allocation via the application of solution concepts (such as the Shapley value or the nucleolus) encounter a problem in games of this form, because the game must first be transformed into some other form. Commonly, this other form is a set function defined over coalitions, such as the von Neumann-Morgenstern characteristic function. Because there are enumerous possible transformations, the question arises as to which one provides the most accurate basis for prediction of payoffs. The laboratory experiment tested three such transformations - the mixed strategy characteristic function, the pure strategy characteristic function, and the Harsanyi threat function. Payoff predictions from two solution concepts (Shapley value, nucleolus) were computed on the basis of each of these transformations, making a total of six theories under test. Results of the study show, in general, that payoff predictions based on the Harsanyi threat function and on the mixed strategy characteristic function were more accurate than those based on the pure strategy characteristic function. The most accurate theories were the Shapley value computed from the Harsanyi function, the nucleolus computed from the Harsanyi function, and the Shapley value computed from the mixed strategy characteristic function. Less accurate were the nucleolus computed from the mixed strategy characteristic function and both the nucleolus and the Shapley value computed from the pure strategy characteristic function.