The skewness of the conditional return distribution plays a significant role in financial theory and practice. This paper examines whether conditional skewness of daily aggregate market returns is predictable and investigates the economic mechanisms underlying this predictability. In both developed and emerging markets, there is strong evidence that lagged returns predict skewness; returns are more negatively skewed following an increase in stock prices and returns are more positively skewed following a decrease in stock prices. The empirical evidence shows that the traditional explanations such as the leverage effect, the volatility feedback effect, the stock bubble model (Blanchard and Watson, 1982), and the fluctuating uncertainty theory (Veronesi, 1999) are not driving the predictability of conditional skewness at the market level. The relation between skewness and lagged returns is more consistent with the Cao, Coval, and Hirshleifer (2002) model. Our findings have implications for future theoretical and empirical models of time-varying market return distributions, optimal asset allocation, and risk management.