We provide evidence that agents have slow-moving beliefs about stock market volatility that lead to initial underreaction to volatility shocks followed by delayed overreaction. These dynamics are mirrored in the VIX and variance risk premiums, which reflect investor expectations about volatility, and are also supported in both surveys and firm-level option prices. We embed these expectations into an asset pricing model and find that the model can account for a number of stylized facts about market returns and return volatility that are difficult to reconcile, including a weak or even negative risk-return trade-off.AGENTS' PERCEPTIONS OF RISK PLAY a critical role in asset pricing models. However, a long literature finds that the empirical risk-return trade-off is weak at best (Glosten, Jagannathan, and Runkle (1993)), despite this trade-off being strong in leading asset pricing models (Moreira and Muir (2017)). 1 This paper proposes a model in which a representative agent has biased, slow-moving expectations about volatility. We show that these expectations help explain a weak relation between risk and returns. We discipline expectations about volatility in the model in three ways: we microfound beliefs based on sticky and extrapolative expectations (which have been shown to be present in many other contexts), we use survey data to directly assess agents' expectations of