Standard preferences in classical economics follow certain "consistency rules" such as transitivity etc.pp. These classical preferences such as expected utility capture only the first two moments of a say normal distribution (expected value and volatility/standard deviation). Lottery-like stocks as defined by Alok Kumar ("Who gambles in the stock market") are characterized by a low price, a high volatility and a certain positive skewness. The latter one is the crucial additional feature as, if you believe in the classical utility paradigm, skewness and other higher moments shouldn't matter for asset allocation. However, if you sympathize with behavioral finance and prospect theory in particular, skewness is taken into account by the decision weights. Barberis and Huang have shown in 2007 ("stocks as lotteries") that given an investor with decision weighting functions instead of classical expectations will perceive a certain amount of lottery-like stocks as favourable compared to a portfolio without them. If all investors think this way, they will put a premium on the price (as markets have to be in an equilibrium), which will result in below-average expected returns from that stock. Thats why it's a paradox: people want to pay a higher price as they perceive those unlikely but high returns (=positive skewness) as more likely as physical probabilities suggest, thus driving up the price and lowering the returns..