A simple three–player model is presented for the evolution of coalitions. The model demonstrates that, under certain conditions, ‘winner’ and ‘loser’ effects both favour coalition formation. Winner effects are defined as an increased probability of winning at time T+ 1, given a victory at time T, whereas loser effects entail an increased probability of losing at time T+1, given a loss at time T. Increasing the strength of loser effects or winner effects, or the strength of an individual's position in the hierarchy, makes coalition formation in general more likely, whereas increasing the costs of giving aid does the opposite. The model does not assume any form of reciprocity, but rather examines whether some form of reciprocity or pseudoreciprocity emerges from the model itself. When either winner or loser effects exist, reciprocal coalition formation (e.g. i helps j against k, and j helps i against k) between β (second–ranked individual) and α (highest–ranked individual) or between α and γ (lowest–ranked individual) was possible, but reciprocal aid–giving between γ and β was never favoured. Thus, we have the counterintuitive result that although a coalition between the two lowest members of a hierarchy against the dominant individual is possible (as selection may favour γ aiding β against α), such a coalition is not predicted to be reciprocal in kind. Interpopulational comparisons examining winner–loser effects and coalition formation would allow for a test of many of the model's most basic predictions. Unfortunately, most work on coalitions has been undertaken in primates, whereas work on winner and loser effects has focused on rodents, and more recently, in fish and birds. Hopefully, the model presented here will spur future work that will look at all of these factors simultaneously in many taxa.