Data on individual trades in prediction markets relating to the 2008 and 2012 US Presidential elections reveal that traders vary enormously in their behavior. This contrasts with the standard prediction-market models, which assume relatively homogeneous participants who differ only in their beliefs and wealth. We show that risk-lovers have particularly strong distortionary effects on market outcomes even when beliefs are symmetrically distributed around the truth. Simulations of a model which allows traders to have different motives and tastes for risk indicate that including such traders produce the market outcomes we observe, such as herding, persistent contrariness, a skewed profits' distribution and favorite-long-shot bias. The attraction of such markets to risk-lovers means that caution must be exercised when using prediction-market prices for forecasting