Mutual Fund Performance with Learning Across Funds

This paper is based on the premise that knowledge about the alphas of one set of funds will influence an investor's beliefs about other funds. This will be true insofar as an investor's expectation about the performance of a fund is partly a belief about the abilities of mutual fund managers as a group and, more generally, a belief about the degree to which financial markets are efficient. We develop a simple framework for incorporating this prior dependence' and find that it can have a substantial impact on the cross-section of posterior beliefs about fund performance as well as asset allocation. Under independence, the maximum posterior mean alpha increases without bound as the number of funds increases and 'extremely large' estimates are randomly observed. This is true even when fund managers have no skill. In contrast, with prior dependence, investors aggregate information across funds to form a general belief about the potential for abnormal performance. Each fund's alpha estimate is shrunk toward the aggregate estimate, mitigating extreme views. An additional implication is that restricting the estimation to surviving funds, a common practice in this literature, imparts an upward bias to the average fund alpha.