We consider an extensive form game with incomplete information which consists of several steps of a kind of regular structure. Since the determination of Nash equilibria of this kind of games is straightforward but gets more and more tedious if the number of steps increases, and since for the purpose of applications to the just mentioned models only those equilibria are interesting which do not imply a premature end of the game at one of the intermediate dead ends, an algorithm is considered the intuitive meaning of which may be described as a learning procedure: If the player, who has only incomplete knowledge of this adversary's type, has reached some information set of the game, he is supposed to know more about the type of the latter one than in the beginning of the game due to the fact that the latter one behaved in some specific way. Working out this idea one can devise a kind of backward induction procedure which, however, contains some recursive elements. It is the purpose of this note to describe this algorithm and furthermore, to show that it leads to all Nash equilibria of the game with the property that all information sets are reached during the course of the game. In addition, it is shown that this algorithm can also be applied to extensive form games with imperfect information which have a similar tree structure as those considered before.    «

We consider an extensive form game with incomplete information which consists of several steps of a kind of regular structure. Since the determination of Nash equilibria of this kind of games is straightforward but gets more and more tedious if the number of steps increases, and since for the purpose of applications to the just mentioned models only those equilibria are interesting which do not imply a premature end of the game at one of the intermediate dead ends, an algorithm is considered the...    »