We study simulations of populations of agents participating in sequences of overlapping English auctions, using three different bidding algorithms. We measure various parameters of the agents’ success, to determine qualities of the underlying bidding algorithms. In particular, we show that a Dynamic Programming approach, in which beliefs regarding the opposition the agent is likely to face are built up on-the-fly, is robust enough with respect to the inaccuracy of its beliefs to outperform a greedy approach right from the moment they both start playing.