During recessions, many macroeconomic variables display higher levels of volatility. We show how introducing an AR(1)-ARCH(1) driving process into the canonical Lucas consumption CAPM framework can account for the empirically observed greater volatility of asset returns during recessions. In particular, agents' joint forecasting of levels and time-varing second moments transforms symmetric-volatility driving processes into asymmetric-volatility endogenous variables. Moreover, numerical examples show that the model can indeed account for the degree of cyclical variation in both bond and equity returns in the U.S. data. Finally, we argue that the underlying mechanism is not specific to financial markets, and has the potential to explain cyclical variation in the volatilities of a wide variety of macroeconomic variables.