I study the cross-section of realized stock option returns and find an economically important source of predictability in the cross-sectional distribution of implied volatility. A zero-cost trading strategy that is long in straddles with a large positive forecast of the change in implied volatility and short in straddles with a large negative forecast produces an economically important and statistically significant average monthly return. The results are robust to different market conditions, to firm risk-characteristics, to various industry groupings, to options liquidity characteristics, and are not explained by linear factor models.