This paper considers two-player repeated games with discounting. Each
player is assumed to form a prediction over a limited horizon while hi
s view about what will come next is only vague. Following Jehiel (1995
) Journal of Economic Theory 67, 497-519, a solution concept is propos
ed in which the players make correct predictions over their horizon of
forecast and the vagueness about what will come next is modelled as a
source of randomness in players' criteria. Applications to the repeat
ed prisoner's dilemma are reported suggesting that in contrast to rece
ived theory it may be easier to sustain cooperation when the players a
re not too patient and it may be easier to sustain cooperation than no
n-cooperation. (C) 1998 Elsevier Science B.V. All rights reserved.