Fictitious play

In game theory, fictitious play is a learning rule first introduced by George W. Brown. In it, each player presumes that the opponents are playing stationary (possibly mixed) strategies. At each round, each player thus best responds to the empirical frequency of play of their opponent. Such a method is of course adequate if the opponent indeed uses a stationary strategy, while it is flawed if the opponent's strategy is non-stationary. The opponent's strategy may for example be conditioned on the fictitious player's last move. Brown first introduced fictitious play as an explanation for Nash equilibrium play. He imagined that a player would "simulate" play of the game in their mind and update their future play based on this simulation; hence the name fictitious play. In terms of current use, the name is a bit of a misnomer, since each play of the game actually occurs. The play is not exactly fictitious. In fictitious play, strict Nash equilibria are absorbing states. That is, if at any time period all the players play a Nash equilibrium, then they will do so for all subsequent rounds. (Fudenberg and Levine 1998, Proposition 2.1) In addition, if fictitious play converges to any distribution, those probabilities correspond to a Nash equilibrium of the underlying game. (Proposition 2.2) Therefore, the interesting question is, under what circumstances does fictitious play converge? The process will converge for a 2-person game if: Both players have only a finite number of strategies and the game is zero sum (Robinson 1951) The game is solvable by iterated elimination of strictly dominated strategies (Nachbar 1990) The game is a potential game (Monderer and Shapley 1996-a,1996-b) The game has generic payoffs and is 2 × N (Berger 2005) Fictitious play does not always converge, however. Shapley (1964) proved that in the game pictured here (a nonzero-sum version of Rock, Paper, Scissors), if the players start by choosing (a, B), the play will cycle indefinitely.