Dynamic multistage games with perfect information are considered. The definition of the game differs from the classical H. Kuhn definition by the presence of vertices in which the chance moves and randomly selects the coalitional partition in the game. This partition remains unchanged until the game finds itself in the next vertex where, the chance move is making the decision to choose the next coalitional partition. The new value for such a game is proposed (the so called PMS-value). This value is computed by using the backward induction procedure for the vertices with a given coalitional partition and more complicated transition procedures in the vertices of the chance. The result is illustrated by an example.