Traditional asset pricing models predict that covariance between prices of different assets should be lower than what we observe in the data. This model generates high covariance within a rational expectations framework by introducing markets for information about asset payoffs. When information is costly, rational investors will not buy information about all assets; they will learn about a subset. Because information production has high fixed costs, competitive producers charge more for low-demand information than for high-demand information. A price that declines in quantity makes investors want to purchase a common subset of information. If investors price many assets using a common subset of information, then a shock to one signal is passed on as a common shock to many asset prices. These common shocks to asset prices generate 'excess covariance.' The cross-sectional and time-series properties of asset price covariance are consistent with this explanation. * lveldkam@stern.nyu.edu, NYU Stern, Economics Department, 44 West 4th St., 7th floor, New York, NY 10012. Thanks to David Backus, Guido Lorenzoni, Thomas Sargent, Martin Schneider, Stijn Van Nieuwerburgh, Bernie Yeung, Gunter Strobl, and seminar participants at NYU, Rutgers, 2004 EFA meetings, Gerzensee summer symposium, and the 2004 SED meetings for helpful comments and conversations. Thanks also to Elif Sisli for her capable research assistance. JEL classification: D82, G14, G12.