When do markets with simple agents fail?

We consider (prediction) markets where myopic agents sequentially interact with an automated market maker. We show a broad negative result: by varying the order of participation, the market’s aggregate prediction can converge to an arbitrary value. In other words, markets may fail to do any meaningful belief aggregation. On the positive side, we show that under a random participation model, steady state prices equal those of the traditional static prediction market model. We discuss applications of our results to the failure of the 1996 Iowa Electronic Market.