We posit a theory that runs counter to how conventional wisdom thinks about analyst bias, that it is the result of distorted incentives by "the system" -especially upstream factors like the analysts' employers. We suggest that analysts are also heavily influenced by what investors believe, the purported victims of analyst bias. We adapt Mullainathan-Shleifer's theory of media bias to build a theory of how analysts cater to what investors believe. The theory also predicts that competition among analysts does not reduce their bias. We provide empirical support for this theory, using an enormous dataset built from over 6.5 million analyst estimates and 42.8 million observations on investor holdings, which we argue is a proxy for what they believe. We use a simultaneousequations model for estimation, with instruments to rule out alternative interpretations of the direction of causality. For additional robustness, we investigate the time series of analyst bias and heterogeneity in investor beliefs from 1987 through 2003. Dickey-Fuller tests show that both have unit roots, but we establish that cointegration hold. Further, we employ a vectorautoregressive model to show Granger-causality between the two.The bias of analyst forecasts has received extensive empirical study, from as early as Mastrapasqua and Bolten (1973) and Brown and Rozeff (1978). A recent survey is in OʹBrien (2003). A large class of studies deals with how analysts' biases might affect stock prices and other aspects of the market. This paper joins an emerging class of studies that seeks to understand the source of the bias. For example, Hong and Kubik (2003) show that analysts who are positively biased get better jobs in the industry. Cowen, Groysberg and Healy (2003) provide evidence that analysts whose employers are more commissionbased, such as pure brokerages, are more bias. Underlying many of these explanations is an upstream source of analyst bias: that of distorted incentives by their employers and the companies covered.In this paper, we ask the seemingly obvious question: might not analysts' biases be a result of their catering to downstream influence -that is, their direct clients, the readers? This paper can also be viewed as a contribution to the already large body of literature on behavorial corporate finance, in which a theme is that firm behavior can be explained by their responses to clients and the market. This theme has been used to explain capital 2 structure (Baker and Wurgler (2002)), dividend payouts (Baker and Wurgler (2003)), and acquisitions (Shleifer and Vishny (2003)).Our "catering" theory, if true, has important practical implications. For example, it means that policy makers have to go beyond reforming incentives and organizational structures inside analysts' institutions, if we were to have any hope of reducing analyst bias. Finally, this research question opens up a range of others. The most important is how readers of analyst reports are in turn influenced by what analysts write.We start with Section I, in which...