In the rational expectations equilibrium of this paper, agents have private information and differing information partitions and therefore assign differing conditional distributions to asset payoffs and other economic variables relevant to their investment choices. Standard asset pricing models typically do not recognize the impact of these differing information partitions, and empirical tests based on these models thus measure asset riskiness in a way that may not be relevant to any of the agents' decisions. I show how this can lead to distorted estimates of investment risk and how it can make the equity premium appear difficult to explain.